Implements Session 2 of the patterns example plan
(livetemplate/livetemplate docs/proposals/patterns.md): Lists & Data
(#8-11) and Search & Filtering (#12-13). Adds 6 new patterns as
self-contained handler+template+state+tests, plus a Session 1 polish
sweep for issues surfaced during manual review.

## New patterns

- **#8 Delete Row** — per-row delete with `lvt-fx:animate="slide"`
  entry animation. Controller holds a process-wide mutex-protected
  in-memory item store so deletions persist across reloads and
  navigation without needing `lvt:"persist"` struct tags. Includes a
  Restore button that repopulates the store.
- **#9 Click To Load** — button-driven pagination over a 25-item
  dataset (3 pages at size 10).
- **#10 Infinite Scroll** — `<div id="scroll-sentinel">`-driven
  pagination over a 100-item dataset, demonstrating the client's
  IntersectionObserver + `loadMorePending` throttle.
- **#11 Value Select** — cascading Make/Model selects via `Change()`
  auto-wiring. Changing Make auto-selects the first Model so the
  cascade is visibly propagated.
- **#12 Active Search** — debounced live search over a 25-contact
  directory, using `Change()` on `<input type="search">`.
- **#13 URL-Preserved Filters** — bookmarkable filter state via URL
  query parameters, read in `Mount()` with graceful fallback for
  unknown values. Validates status ∈ {all, active, completed} and
  sort ∈ {name, date}.

All 6 patterns follow the Session 1 conventions: per-category
state_*.go and handlers_*.go files, `{{define "content"}}` templates
extending layout.tmpl, `data-key` on range items, chromedp E2E tests
with Initial_Load / UI_Standards / Visual_Check subtests plus
action-specific coverage.

## Session 1 polish sweep

- **Edit Row**: dropped hidden-input ID pattern in favor of button
  `value` attribute (`<button name="edit" value="{{.ID}}">` →
  `ctx.GetString("value")`), matching the documented Tier 1 pattern
  in progressive-complexity-reference.md. Same change applied to
  Delete Row's new template.
- **Click To Edit**: wrapped view-mode Edit button in a form for Tier 1
  no-JS compliance (fixes Session 1 tracker issue #66).
- **Bulk Update**: fixed spurious "Updated N user(s)" flash always
  reporting the total user count. Now counts actual checkbox changes
  and shows a "No changes" info flash when none changed. Added
  `{{.lvt.FlashTag "info"}}` to the template since only success was
  previously rendered.

## Cross-handler navigation regression test

Added `TestCrossHandlerNavigation/Index_To_Delete_Row_No_Stale_Dom` to
lock in DOM structure (article/h4/row counts) after index→pattern
navigation, catching treeRenderer state-leak bugs that pre-existing
string-contains assertions miss.

## Infrastructure

- `runStandardSubtests(t, ctx, pico, desc)` helper consolidates 12
  copy-pasted `UI_Standards` + `Visual_Check` subtest pairs across
  Session 1 and Session 2 tests. Delete Row retains its hand-rolled
  `UI_Standards` because it needs an animation-wait precheck.
- Data layer cleanup: `buildItemDataset(n, prefix)` shared builder for
  `listDataset` (25) and `infiniteScrollDataset` (100); `slices.Clone`
  everywhere replacing manual make+copy idioms; `slices.Sorted(maps.Keys)`
  for `getCarMakes`; hoisted contactDirectory to package-level var.

## Dependency

CI runs against `@latest` client from CDN. Once
github.com/livetemplate/client#69 (cross-handler nav + infinite scroll
race + animation fixes) is merged and released, CI here will pick up
the new client and pass. Until then, expect test failures on
`TestCrossHandlerNavigation/Index_To_Delete_Row_No_Stale_Dom` and
possibly `TestInfiniteScroll/Auto_Paginate_To_End` against the stale
client.

Tested locally with `LVT_LOCAL_CLIENT` pointing at the client PR's
build: 15 tests pass, ~55s runtime.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* feat: patterns example Session 2 (patterns #8-13) + polish sweep

Implements Session 2 of the patterns example plan
(livetemplate/livetemplate docs/proposals/patterns.md): Lists & Data
(#8-11) and Search & Filtering (#12-13). Adds 6 new patterns as
self-contained handler+template+state+tests, plus a Session 1 polish
sweep for issues surfaced during manual review.

## New patterns

- **#8 Delete Row** — per-row delete with `lvt-fx:animate="slide"`
  entry animation. Controller holds a process-wide mutex-protected
  in-memory item store so deletions persist across reloads and
  navigation without needing `lvt:"persist"` struct tags. Includes a
  Restore button that repopulates the store.
- **#9 Click To Load** — button-driven pagination over a 25-item
  dataset (3 pages at size 10).
- **#10 Infinite Scroll** — `<div id="scroll-sentinel">`-driven
  pagination over a 100-item dataset, demonstrating the client's
  IntersectionObserver + `loadMorePending` throttle.
- **#11 Value Select** — cascading Make/Model selects via `Change()`
  auto-wiring. Changing Make auto-selects the first Model so the
  cascade is visibly propagated.
- **#12 Active Search** — debounced live search over a 25-contact
  directory, using `Change()` on `<input type="search">`.
- **#13 URL-Preserved Filters** — bookmarkable filter state via URL
  query parameters, read in `Mount()` with graceful fallback for
  unknown values. Validates status ∈ {all, active, completed} and
  sort ∈ {name, date}.

All 6 patterns follow the Session 1 conventions: per-category
state_*.go and handlers_*.go files, `{{define "content"}}` templates
extending layout.tmpl, `data-key` on range items, chromedp E2E tests
with Initial_Load / UI_Standards / Visual_Check subtests plus
action-specific coverage.

## Session 1 polish sweep

- **Edit Row**: dropped hidden-input ID pattern in favor of button
  `value` attribute (`<button name="edit" value="{{.ID}}">` →
  `ctx.GetString("value")`), matching the documented Tier 1 pattern
  in progressive-complexity-reference.md. Same change applied to
  Delete Row's new template.
- **Click To Edit**: wrapped view-mode Edit button in a form for Tier 1
  no-JS compliance (fixes Session 1 tracker issue #66).
- **Bulk Update**: fixed spurious "Updated N user(s)" flash always
  reporting the total user count. Now counts actual checkbox changes
  and shows a "No changes" info flash when none changed. Added
  `{{.lvt.FlashTag "info"}}` to the template since only success was
  previously rendered.

## Cross-handler navigation regression test

Added `TestCrossHandlerNavigation/Index_To_Delete_Row_No_Stale_Dom` to
lock in DOM structure (article/h4/row counts) after index→pattern
navigation, catching treeRenderer state-leak bugs that pre-existing
string-contains assertions miss.

## Infrastructure

- `runStandardSubtests(t, ctx, pico, desc)` helper consolidates 12
  copy-pasted `UI_Standards` + `Visual_Check` subtest pairs across
  Session 1 and Session 2 tests. Delete Row retains its hand-rolled
  `UI_Standards` because it needs an animation-wait precheck.
- Data layer cleanup: `buildItemDataset(n, prefix)` shared builder for
  `listDataset` (25) and `infiniteScrollDataset` (100); `slices.Clone`
  everywhere replacing manual make+copy idioms; `slices.Sorted(maps.Keys)`
  for `getCarMakes`; hoisted contactDirectory to package-level var.

## Dependency

CI runs against `@latest` client from CDN. Once
github.com/livetemplate/client#69 (cross-handler nav + infinite scroll
race + animation fixes) is merged and released, CI here will pick up
the new client and pass. Until then, expect test failures on
`TestCrossHandlerNavigation/Index_To_Delete_Row_No_Stale_Dom` and
possibly `TestInfiniteScroll/Auto_Paginate_To_End` against the stale
client.

Tested locally with `LVT_LOCAL_CLIENT` pointing at the client PR's
build: 15 tests pass, ~55s runtime.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix(patterns/test): harden TestActiveSearch/Clear_Query_Restores_All for CI

CI-specific flake: the test passes locally in ~0.4s but hit the 5s
timeout in CI consistently across both runs on this PR. Root cause is
resource pressure — the failing assertion still showed the previous
"Chen" query state with 1 row, indicating the dispatched `input` event
either never reached the Change auto-wirer debounce or the server
response never made it back within the tight timeout.

Changes:
- Bump WaitForCount timeout from 5s to 10s so CI has headroom under
  orphan-process load.
- Add WaitForWebSocketReady(5s) before the dispatch to ensure the WS
  is fully re-stabilized after Filter_To_Single_Result's SendKeys
  debounce cascade finished.
- Focus the input explicitly before dispatching (some browsers ignore
  events on unfocused inputs under stress).
- Defensively dispatch both `input` AND `change` events — the
  auto-wirer listens on `input` for text inputs, but this is cheap
  insurance against any future event-filter change.

Local: still completes in 0.48s. No regression on the happy path.

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Functional fixes from Claude bot + Copilot:

- Pattern numbering collision: my new headers labeled the Sortable
  example as "Pattern #12", but Active Search already had that number.
  Drop the global numbering from the new code rather than cascade-
  renumbering 25+ comments through unrelated patterns.
- Reorder_DragForward had a comment that contradicted the assertion
  ("task-2, task-3, task-1, ..." vs the actual "task-2, task-1,
  task-3, ..."). Comment was wrong; rewritten to match the algorithm.
- Reorder_DragBackward had been stateful — relied on DragForward
  having run first. Now resets to initial order, which also exposed
  a bug in MY expected order (insert-before-target with srcIdx > tgtIdx
  gives [task-1, task-6, task-2, ...], not [task-6, ...]). Fixed.
- SelfDrop_NoOp ignored the chromedp.Run error from the in-page poll
  (`_ = chromedp.Run(...)`), and the async IIFE returned an unawaited
  Promise that chromedp couldn't unmarshal as bool — silently passing.
  Switched to a Go-side time.Sleep + before/after compare. Errors are
  now captured and t.Fatal'd.
- Reorder subtests no longer depend on each other: each now starts
  with a Reset-to-initial via a shared resetToInitial Tasks block.

Template / docs:

- Hgroup subtitle: replace the "diff engine sends only ['o', newKeys]"
  implementation-detail prose with user-facing copy. Drop the false
  "broadcasts to other tabs" claim — the demo persists across reloads
  and is visible to other tabs on next refresh, but doesn't push live
  updates (no Sync/BroadcastAction).
- Add aria-label="Reorderable task list" to the <ul>.
- Drop what-comments per CLAUDE.md (SortableState struct comment,
  initialSortableItems doc).
- Update SortableController doc to remove the stale "broadcast" claim.

Skipped: aria-live announcement region (separate scope, would need a
server-side state field for the announcement text).

TestSortable still passes (5/5 subtests).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Functional fixes from Claude bot:

- Reorder() guard returns now populate state.Items from snapshot.
  Previously the early-exit paths returned `state` as-is, which the
  framework may have populated with a stale Items slice — the bug
  could overwrite the rendered list with stale data on cross-app
  drags or unknown keys. Snapshot-on-every-return is now the
  invariant, matching how Mount() already worked.
- Reset_RestoresInitialOrder was trivially passing because the
  preceding subtest (SelfDrop_NoOp) calls resetToInitial first and
  the self-drop is a no-op, so by the time Reset_RestoresInitialOrder
  ran the list was already in initial order. Now scrambles the list
  first (drag task-3 onto task-1), then asserts Reset reverts it.
- Subtitle copy: drop "by its handle" (the entire <li> is draggable,
  not just the hamburger glyph — the glyph is decorative). Add
  "(mouse only — no keyboard reorder path in this demo)" to set
  expectations honestly.
- Drop the SortableItem.Key what-comment per CLAUDE.md.

Skipped:
- Renumbering existing patterns (Active Search "#12", URL-Preserved
  Filters "#13"): those labels are pre-existing legacy code outside
  this PR's scope. The numbering scheme is now historical-not-ordinal,
  acknowledged by Sortable having no number.
- Bumping time.Sleep to 1s in SelfDrop_NoOp: 500ms is sufficient in
  practice; longer waits add CI time without clear benefit.
- Refactoring the template to drag-by-handle instead of drag-by-row:
  whole-row draggable is the simpler pedagogical demo. Subtitle copy
  is fixed to no longer mislead users about handle-only behavior.

TestSortable still passes (5/5 subtests).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* feat(patterns): sortable list with HTML5 drag-and-drop

Adds the Sortable List pattern (Lists & Data #12), demonstrating the
new lvt-on:dragstart / lvt-on:dragover / lvt-on:drop event support
shipped in @livetemplate/client (livetemplate/client#101).

The template is a plain <ul> of draggable <li>s, each carrying a
data-key. The client auto-stashes the dragged item's data-key into
DataTransfer on dragstart and lifts it back out on drop, so the
controller's Reorder() method receives a {dragSourceKey, dragTargetKey}
pair without any custom JavaScript. lvt-mod:throttle="100" tames
dragover's high frequency. The diff engine emits only ["o", newKeys]
on the wire — no full re-render.

SortableController follows DeleteRowController's process-wide in-memory
shared-state pattern so reorders persist across reloads and across
tabs (multi-tab broadcast story works for free). A Reset Order button
restores the initial ordering.

E2E coverage exercises Reorder forward, backward, the self-drop no-op
guard, and Reset. Drag is simulated by dispatching real DragEvent
objects with a shared DataTransfer via chromedp.Evaluate — this hits
the production client delegation pipeline rather than bypassing it
via liveTemplateClient.send(), but works around the unreliable HTML5
DnD synthesis through CDP Input.dispatchMouseEvent in headless Docker
Chrome.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(patterns): use empty-action drag bindings (marker pattern)

The Sortable pattern was binding lvt-on:dragstart and lvt-on:dragover
to "reorder", which caused one no-op Reorder call per dragstart plus
~10/sec throttled no-op calls per dragover tick. Drop is the only
event that carries meaningful keys.

Switch to the marker pattern (empty action value) introduced in
@livetemplate/client #101 follow-up: the client still runs the
spec-mandated side-effects (preventDefault, dataTransfer.setData) but
skips the WS round-trip. Net result: exactly one Reorder call per
drop gesture instead of dozens of no-ops.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(patterns): tighten Sortable controller + trim test prose

Code-review pass:

- SortableController.Reorder: single-pass key scan with early break
  once both indices are found (academic for 6 items, real if a future
  example uses a longer list).
- SortableController.Reset: hold the lock across the swap+clone instead
  of releasing and re-acquiring via snapshot().
- Comments: drop the "mirrors DeleteRowController" narration and the
  Reorder algorithm prose that just restated the code; keep the
  client-contract description for ctx field names. Trim the
  simulateDrag and SelfDrop_NoOp test rationale to one line each.

No behavior change. TestSortable still passes (3.4s).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: address bot review on PR #84

Functional fixes from Claude bot + Copilot:

- Pattern numbering collision: my new headers labeled the Sortable
  example as "Pattern #12", but Active Search already had that number.
  Drop the global numbering from the new code rather than cascade-
  renumbering 25+ comments through unrelated patterns.
- Reorder_DragForward had a comment that contradicted the assertion
  ("task-2, task-3, task-1, ..." vs the actual "task-2, task-1,
  task-3, ..."). Comment was wrong; rewritten to match the algorithm.
- Reorder_DragBackward had been stateful — relied on DragForward
  having run first. Now resets to initial order, which also exposed
  a bug in MY expected order (insert-before-target with srcIdx > tgtIdx
  gives [task-1, task-6, task-2, ...], not [task-6, ...]). Fixed.
- SelfDrop_NoOp ignored the chromedp.Run error from the in-page poll
  (`_ = chromedp.Run(...)`), and the async IIFE returned an unawaited
  Promise that chromedp couldn't unmarshal as bool — silently passing.
  Switched to a Go-side time.Sleep + before/after compare. Errors are
  now captured and t.Fatal'd.
- Reorder subtests no longer depend on each other: each now starts
  with a Reset-to-initial via a shared resetToInitial Tasks block.

Template / docs:

- Hgroup subtitle: replace the "diff engine sends only ['o', newKeys]"
  implementation-detail prose with user-facing copy. Drop the false
  "broadcasts to other tabs" claim — the demo persists across reloads
  and is visible to other tabs on next refresh, but doesn't push live
  updates (no Sync/BroadcastAction).
- Add aria-label="Reorderable task list" to the <ul>.
- Drop what-comments per CLAUDE.md (SortableState struct comment,
  initialSortableItems doc).
- Update SortableController doc to remove the stale "broadcast" claim.

Skipped: aria-live announcement region (separate scope, would need a
server-side state field for the announcement text).

TestSortable still passes (5/5 subtests).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: address round-2 bot review on PR #84

Functional fixes from Claude bot:

- Reorder() guard returns now populate state.Items from snapshot.
  Previously the early-exit paths returned `state` as-is, which the
  framework may have populated with a stale Items slice — the bug
  could overwrite the rendered list with stale data on cross-app
  drags or unknown keys. Snapshot-on-every-return is now the
  invariant, matching how Mount() already worked.
- Reset_RestoresInitialOrder was trivially passing because the
  preceding subtest (SelfDrop_NoOp) calls resetToInitial first and
  the self-drop is a no-op, so by the time Reset_RestoresInitialOrder
  ran the list was already in initial order. Now scrambles the list
  first (drag task-3 onto task-1), then asserts Reset reverts it.
- Subtitle copy: drop "by its handle" (the entire <li> is draggable,
  not just the hamburger glyph — the glyph is decorative). Add
  "(mouse only — no keyboard reorder path in this demo)" to set
  expectations honestly.
- Drop the SortableItem.Key what-comment per CLAUDE.md.

Skipped:
- Renumbering existing patterns (Active Search "#12", URL-Preserved
  Filters "#13"): those labels are pre-existing legacy code outside
  this PR's scope. The numbering scheme is now historical-not-ordinal,
  acknowledged by Sortable having no number.
- Bumping time.Sleep to 1s in SelfDrop_NoOp: 500ms is sufficient in
  practice; longer waits add CI time without clear benefit.
- Refactoring the template to drag-by-handle instead of drag-by-row:
  whole-row draggable is the simpler pedagogical demo. Subtitle copy
  is fixed to no longer mislead users about handle-only behavior.

TestSortable still passes (5/5 subtests).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: address round-3 bot review on PR #84

- Subtitle: also document that there's no visual drop-zone highlight
  in this v1 demo (the framework doesn't currently expose a hover
  class hook for drag targets, and adding one would require client
  changes outside this example's scope).
- SelfDrop_NoOp time.Sleep: bump to 1s to give loaded CI runners
  headroom for any spurious round-trip to surface in the assertion.
- Reorder() docstring: trim the second half that walked through the
  algorithm; keep the non-obvious invariant about not trusting
  framework-provided state on early-exit paths.

Skipped:
- Adding a Reorders counter field to SortableState as a positive
  assertion replacement for time.Sleep — that pollutes the wire
  format with a test-only field and adds non-trivial state-design
  complexity for marginal robustness improvement.

TestSortable still passes (5/5 subtests).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: address round-4 bot review on PR #84

- Add aria-description on the <ul> so AT users get an explicit notice
  about the missing keyboard reorder path, instead of hitting a silent
  dead-end on draggable items.
- Tighten the Reorder() doc comment to name the source of the ctx keys
  (the livetemplate/client drag handler reading data-key on the
  dragstart and drop targets).

Skipped:
- dragleave/dragend abort guard: moot — the client only fires the
  Reorder action on actual drop events. Aborted drags (release outside
  the list, ESC) don't fire drop on a registered lvt-on:drop element,
  so Reorder isn't called at all.
- Glyph choice (☰ hamburger vs ⋮ vertical ellipsis): both are commonly
  used as drag-handle indicators. Bikeshedding.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix: address round-5 bot review on PR #84

- aria-description (ARIA 1.3, uneven AT support) → aria-describedby
  pointing to a visually-hidden description element. Reliably
  surfaces the keyboard-reorder limitation to screen readers.
- Trim multi-line comment blocks on SortableController and Reorder
  per CLAUDE.md "one short line max" rule. Keep the load-bearing
  WHY (snapshot-on-every-return invariant) on a single line.

Skipped:
- dragend-without-drop edge case: dataTransfer is per-event, no
  persistent client state to dangle.
- Replacing time.Sleep(1s) with an echo-ping pattern: adds complexity
  for marginal robustness; declined twice already.
- "Subtitle leaks 'diff engine' implementation detail": stale read —
  that prose was removed in commit ff7e36c (round 1).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>