♻️ Migrate operational review documents from Show to built-in <If>

[taras]

Closes #219.

Why

The review library shipped its own Show.md conditional before the engine had
one. <If> is now a native structural directive, so the seven operational
review documents that still invoked Show duplicated engine behavior and let
the review system drift from the language its own specification documents.

The release-order blocker recorded in the original issue is gone:
.github/workflows/review.yml prepares and builds the checked-out revision and
runs ./dist/xmd, so no release or workflow change is involved.

What changes

Before:

<Show when={props.pr.stats.totalChanges > 20}>
...
</Show>

After:

<If condition={props.pr.stats.totalChanges > 20}>
...
</If>

Thirteen blocks across seven documents, renamed in place. Bodies, indentation,
line wrapping and surrounding blank lines are unchanged, and
.reviews/components/Show.md is deleted. Finding.md keeps its public when
prop; only its internal element becomes <If condition={props.when}>.

Every migrated condition already evaluates to a boolean and none passed
Show's optional fallback, so no <Else> branch is introduced and #258's
possible truthiness change stays independent of this PR.

How it works

<If> name → expansion engine (before filesystem resolution) → selected branch only

<If> is native, so it is handled by expansion and never resolved from
componentDirs. That is the property the new tests exercise directly: they run
the real operational sources with no If.md or Show.md anywhere in the stub
filesystem.

Review guide

Start with: scripts/tests/review-infrastructure.test.ts

Then review:

1. The seven .reviews/ documents — a pure tag/prop rename.
2. retires the Show component from the review library — the static guard.
3. The five operational tests that execute the real documents.
4. packages/core/tests/unused-in-diff.test.ts — fixture removal plus an exact
   output assertion.
5. specs/code-review-agent-spec.md §5.10.

Look carefully at: the exact trimmed-equality assertions. They intentionally
lock the blank lines the migrated tags surround, including the five blank lines
OxlintSummary's two suppressed blocks leave behind.

What must stay true

• Show is gone from the review library — enforced by deletion and checked by
  retires the Show component from the review library, which walks
  .reviews/**/*.md for an opening or closing Show element, rejects a
  Show.md path, and scans the focused test and the code-review spec. The guard
  asserts the walk found Finding.md first, so it cannot pass vacuously.
• The structural name wins before filesystem component resolution — checked by
  every operational test, none of which supplies an If or Show component.
• An unselected branch does no provider work — checked by
  skips the ExtraneousCodePolicy sample below the review threshold.
• The scan stays scoped. packages/core/tests/expression-props.test.ts uses
  Show as an arbitrary component name and is deliberately untouched.

How to verify it

• retires the Show component from the review library fails if any .reviews/
  document, the focused test, or the spec still carries Show.
• renders Finding's selected icon and message and suppresses its false case
  proves exact branch output (🔴 Broken contract.) and an exactly empty false
  case; it fails on a whitespace change inside the migrated block.
• renders OxlintSummary's clean section and its unavailable warning and
  renders RepoCleanupPolicy's clean section without running either branch
  prove the clean-section and warning paths byte-exactly.
• suppresses ReleaseSpecWarning for ordinary files and warns on release changes proves both directions of its release-config condition.
• expands UnusedInDiff and CommentReview to nothing without an If component
  proves CommentReview's typed data component executes (it fetches the PR's
  comments) while both captures and both trailing branches stay empty.
• skips the ExtraneousCodePolicy sample below the review threshold installs a
  counting Sample provider: 20 changes make zero calls and render the clean
  section, 21 changes make exactly one call and surface the provider result. The
  selected control is what keeps the zero-call case from being vacuous.
• renders the disclosure with symbol, location, count and reason in
  unused-in-diff.test.ts is now exact trimmed equality over the whole
  disclosure, so summary/table whitespace is locked.

Each new regression was mutation-checked: reverting Finding.md to <Show>
(with Show.md restored), reverting the spec example, widening the
ExtraneousCodePolicy threshold to >= 20, padding Finding's rendered line,
and removing a blank line in UnusedInDiff's table each fail the intended test
and nothing else.

Scope

Included

• The thirteen-block rename in seven operational documents and the deletion of
  Show.md.
• Focused regressions in the two suites that own this boundary.
• §5.10 of the code-review-agent specification.

Intentionally unchanged

• <If> implementation, syntax, validation, and truthiness semantics (Coerce the <If> condition with JS truthiness #258).
• specs/executable-mdx-spec.md, which already specifies native <If>/<Else>,
  filesystem precedence, and selected-branch non-execution.
• .github/workflows/review.yml and every other workflow.
• packages/core/tests/expression-props.test.ts, where Show is an arbitrary
  component-resolution fixture rather than a review-library claim.
• UnusedInDiff's selected-disclosure and suppression coverage stays in its own
  suite, because it crosses eval, <Each>, and filesystem fixture boundaries;
  the infrastructure matrix only uses its empty case.

Generated or mechanical changes

• The seven .reviews/ documents contain no intended behavior change. The only
  cosmetic consequence is in OxlintSummary.md, whose wrapped condition keeps
  its authored continuation indentation and therefore no longer aligns under the
  longer condition={ prefix.

Verification

All required local checks are green on db78a89:

deno task lint                     0 errors, formatting clean
deno task check                    no errors
deno task check:jsr                Success Dry run complete
deno task test <3 suites>          11 passed (77 steps), 0 failed
pnpm exec tsx --tsconfig tsconfig.node.json --test <2 suites>   23 pass, 0 fail
bun test <2 suites>                23 pass, 0 fail
git diff --check                   clean

verify:clean and the full per-runtime suites were not run: this change touches
no dependency state, build purity, test discovery, shared setup, or runtime
adapter. CI owns the exhaustive corpus.

Risks and limitations

• Local review integration evidence: not run. deno task review:local
  requires a reachable Ollama at localhost:11434 serving
  qwen3:30b-a3b plus GITHUB_TOKEN; neither is available in this
  environment. This PR's own review workflow builds the checked-out source and
  runs ./dist/xmd, so its marked review comment is the end-to-end evidence and
  is a merge blocker if it fails, errors, or posts nothing substantive.

Scope confirmation

• Every changed file supports the purpose described above.
• Unrelated cleanup and formatting changes are excluded.
• Generated or mechanical changes are clearly identified.
• The description matches the final diff and test results.

[github-actions]

PR #406: ♻️ Migrate operational review documents from Show to built-in

11 files, +339 / -64

Scope

🟡 403 lines changed. PRs under 400 receive more thorough review.

Structural

✅ No structural bloat detected.

Slop

✅ Slop indicators look low.

Static Analysis

✅ Oxlint found no issues.

Correctness

No extraneous code patterns detected.

[taras]

Integration evidence

PR review workflow (required evidence). Run
31308999146,
revision db78a89. It ran deno task setup, deno task build, and
./dist/xmd run .reviews/ReviewPR.md, so the review was produced by this PR's
own engine and its own migrated documents. The posted <!-- xmd-review -->
comment is substantive and contains no ERROR marker.

Two of its sections are direct end-to-end proof of the migration:

• ### Static Analysis is OxlintSummary.md with all three <If> blocks
  false, rendering only ReviewSection's clean text.
• ### Correctness is ExtraneousCodePolicy.md with its <If> selected
  (403 changes > 20): the branch reached <Sample>, the provider ran, and the
  model returned No extraneous code patterns detected.

So the same document supplies both directions of the migrated condition in CI —
suppressed locally in the focused regression, selected here against a real
provider.

Local review (deno task review:local): not run. .reviews/ReviewPR.local.md
needs Ollama serving qwen3:30b-a3b at localhost:11434 plus GITHUB_TOKEN;
neither was available in the implementation environment. Recorded as a missing
prerequisite, not a test failure.

CI: all checks pass — green, lint, jsr, site, smoke, review,
test-deno, test-node, test-bun, deploy. composability skips by design
(it runs on main only).