chore(tauri-commands): remove the last 18 unreachable registrations

[Chloe-JY]

Problem

A sweep of the 1,019 entries in src-tauri/src/commands/handler_list.inc found
145 commands with no caller in src/, tests/, or src-tauri/capabilities/.
Most are simply dead and are removed elsewhere. 18 are not, and each needs a
different fix than deletion:

15 have no frontend caller but are called from Rust. Registering them is
what is dead, not the code:

Command	Called from
internal_browser_get_state / _click / _input / _select / _scroll / _show_mask / _hide_mask / _clean_up / _is_ready (9)	agent-core tools/impls/web/control_internal_browser.rs
list_internal_browser_targets	same, resolve_ready_target / execute_list
get_active_internal_browser_state	list_internal_browser_targets
cli_agent_run	agent_sessions::cli::agent_core_bridge
get_webview_cookies	Cursor session-token capture
get_watch_status	git::watch::commands::watch_repos
convert_patch_to_unified	es_convert_patch_to_diff, extract_apply_patch

Deleting these would break the agent's browser-control tool and CLI session
startup — a real hazard, because they look exactly like the dead ones from the
frontend's point of view. What is actually wrong is that each carries a
#[tauri::command] attribute and a registry entry that expose an IPC endpoint
nothing invokes.

3 are reached only by their own tests: get_git_repo_info and
calculate_ahead_behind (crates/git/src/bundle.rs) and
compute_diff_with_hunks (crates/perf-utils). Registered, never called by
production code in either language, and green in CI forever because the only
thing exercising them is the test written alongside them.

Solution

For the 15: drop the handler_list.inc entry and the #[tauri::command]
attribute; keep the function. Each takes ordinary parameters (AppHandle,
String, request structs) — none takes tauri::State, so removing the
attribute is signature-neutral and the existing Rust callers are unaffected.
convert_patch_to_unified used the #[command] short form; the now-unused
use tauri::command import goes with it.

For the 3: delete the function, its registration, and the tests that were
its only caller — 4 tests in crates/git/src/tests/bundle_tests.rs and 7 in
crates/perf-utils/src/tests/diff_patch_tests.rs. bundle_tests.rs keeps its
git_commit regression test, which covers live code.

Removing compute_diff_with_hunks orphaned the whole
crates/perf-utils/src/diff_patch/structured.rs module (305 lines:
compute_structured_diff_internal, group_into_hunks, generate_split_rows) —
it had exactly one consumer. In crates/git/src/bundle.rs, deleting
get_git_repo_info and calculate_ahead_behind left has_uncommitted_changes,
get_current_branch, get_head_sha and count_commits with no callers, and
the GitRepoInfo / AheadBehindStatus DTOs they returned with no referents.
All are removed. The live compute_structured_diff entry point in
diff_patch/mod.rs and bundle.rs's git_commit are untouched.

Resulting invariant: an entry in handler_list.inc means the frontend can call
it. Rust-internal helpers are plain functions, and code whose only voter is its
own test is gone rather than pretending to be covered.

Potential risks

• The 15 are the risky half of this change, and nothing about them should
  change at runtime. Their bodies, signatures, and callers are identical; only
  the attribute and the registry line are gone. The failure mode to watch for is
  a frontend caller this sweep did not see — but that is the same evidence that
  put them in this bucket, and the agent-side callers are Rust, not IPC.
• Agent browser control is the blast radius if that evidence is wrong. If any
  internal_browser_* command turns out to be invoked from the webview after
  all, the agent's web tool would start failing at runtime with "command not
  found". Verified against every .ts/.tsx file in the repository plus
  tests/e2e/ and src-tauri/capabilities/default.json, and the repository's
  only two indirect invoke() sites both pass literal constants.
• Deleted tests were not covering anything else. All 11 exercise functions
  deleted in the same commit. This lowers the test count without lowering
  coverage of live code — but it does mean crates/perf-utils loses its
  hunk-grouping tests along with the hunk-grouping code.
• structured.rs is a real algorithm, not scaffolding. It implements hunk
  grouping and split-row generation over similar. It is deleted because its
  single consumer is gone, not because it was wrong. If a future diff view wants
  hunks, this commit is where to recover it.
• Rollback: revert this commit. No schema, migration, persisted format, or
  public API is touched.

Verification

• cargo check --workspace --all-targets — clean, zero warnings. Run with a
  private CARGO_TARGET_DIR: two ORGII worktrees resolve this package to the
  same build-script OUT_DIR, so a shared target dir verifies the wrong
  worktree's handler_list.inc.
• cargo clippy --workspace --all-targets -- -D warnings — exit 0.
• cargo test -p browser -p git -p perf_utils -p org2 — 1,324 passed, 0
  failed, 5 ignored (exit 0), re-run after the final rebase.
• Rebased onto develop as refactor(search): remove the semantic search stack #1172, build(dev-server): add rspack dev server, default on macos #1174, feat(appearance): add light/dark skins with per-variant accent #1176, chore(tauri-commands): remove 114 registered commands with no caller #1177, chore(unused-exports): harvest the dirs the check silently excluded #1178, fix(database): serialize the tests that race on the global pool #1180 and
  fix(agent-core): lock ORGII_HOME while the workspace test reads it #1181 landed underneath it. handler_list.inc and bundle.rs were both
  re-derived from the new develop by re-running the removal scripts rather
  than hand-merged, so the diff is exactly the 18 registration lines plus the
  intended function removals. The unrelated db_browser/tempfile lockfile
  drift each rebase pulled in was dropped every time, so Cargo.lock is
  untouched.
• The bundle.rs cascade only appeared once chore(tauri-commands): remove 114 registered commands with no caller #1177 merged. With both PRs'
  removals in the same tree, get_current_branch / get_head_sha /
  count_commits lost their last callers and clippy failed on
  -D warnings. They are removed here, and the eight bundle DTOs that chore(tauri-commands): remove 114 registered commands with no caller #1177
  alone orphaned (GitBundleResult, BundleProgress, ApplyBundleResult,
  PushBundleResult, CloudMergeResult, GetRepoBranchesResult, BranchName,
  LocalCommitInfo) are deliberately left — they are not this PR's
  responsibility and belong in a follow-up.
• Two pre-existing flakes on process-global state (database pool,
  agent_core ORGII_HOME) are fixed in fix(database): serialize the tests that race on the global pool #1180 and fix(agent-core): lock ORGII_HOME while the workspace test reads it #1181; neither is caused
  by this PR and neither is folded in here.
• Classification evidence. For each of the 145 no-frontend-caller commands,
  every Rust reference was located, definitions / use statements (including
  multi-line use {...} blocks) / comments were discarded, each remaining site
  was resolved to its enclosing fn, and the result was iterated to a fixpoint
  so a command whose only caller is another dead command counts as dead. The 15
  in this PR are the ones with a caller that survives that fixpoint; the 3 are
  the ones whose only surviving caller is a test.
• Signature check before removing the attribute. Each of the 15 was inspected
  for tauri::State parameters, which only resolve inside a command. None has
  one, so the attribute is removable without touching call sites.
• Cascade closed. After deleting the 3, cargo check surfaced
  compute_structured_diff_internal, group_into_hunks, generate_split_rows,
  and has_uncommitted_changes as newly unused; each was confirmed against
  origin/develop to be reachable only from a function deleted here before
  being removed. Re-checked to zero warnings afterwards.
• Not run: no TypeScript changes, so pnpm typecheck / vitest / Playwright
  were not run. No UI change, so no screenshots.
• Pre-commit hook trailer is absent. Committed from a git worktree, where
  .husky/_/husky.sh is gitignored and therefore never created by
  git worktree add, and node_modules is not installed. That trailer is
  structurally always missing in a worktree and carries no signal here; the
  check it would have run for a Rust diff is cargo clippy, run manually above.