query: add COMMIT entry to corpus parity _FIELD_TO_KIND (#81)

[IceRhymers]

Refs #81

What & why

tests/unit/test_query_corpus_parity.py's _FIELD_TO_KIND map was missing a "commit": TokenKind.COMMIT entry, even though app/query/parser.py supports commit: as a first-class field. Any commit: corpus entry added to the shared Python/TS parity corpus would trigger a latent KeyError('commit') in test_safe_entries_tokenize_to_the_corpus_atoms — this closes that gap and proves it with a real corpus entry rather than leaving it as a theoretical hole.

Changes

• tests/unit/test_query_corpus_parity.py — add "commit": TokenKind.COMMIT, to _FIELD_TO_KIND, positioned after "branch" to match parser.py's _FIELD_KINDS ordering.
• webui/frontend/src/utils/queryModel.corpus.json — append two safe commit: entries (lowercase hex hashes, to stay valid under Python's hash-lowering + hex-validation and TS's raw-case, any-bareword acceptance):
  • commit:abc1234 (single atom)
  • repo:acme commit:abc1234 foo (three atoms, exercises commit: alongside other fields)

Test-only change; no production code touched.

TDD proof

• Step A (corpus entries added, map entry not yet added): make test → KeyError: 'commit' in test_safe_entries_tokenize_to_the_corpus_atoms[commit:abc1234] and [repo:acme commit:abc1234 foo] — 2 failed / 1059 passed. Reproduces the issue.
• Step B (map entry added): parity suite alone → 88 passed (84 baseline + 2 new cases × 2 parametrized test functions).
• Step C: make webui-test → TS recognize() classifies both new entries safe with identical atoms — 209 passed (209), 12 files.

Test plan (actual output)

• make lint — pass: ruff check clean, ruff format --check (130 files already formatted), mypy app indexer webui — no issues in 35 source files.
• make test — pass: 1061 passed, 237 deselected (parity suite alone: 88 passed).
• make webui-test — pass: Test Files 12 passed (12), Tests 209 passed (209).
• make test-integration — N/A: test-only change, no DB surface touched. Integration coverage (ci-lakebase.yml) is Lakebase-only and unprovisioned repo-wide; deferred to CI, not silently skipped.

Acceptance mapping

• AC1 (map entry exists & correct): _FIELD_TO_KIND["commit"] present, ordering matches parser.py's _FIELD_KINDS.
• AC2 (safe commit: atom exercised through atom-matching path): Step A red (KeyError) → Step B green is the proof.
• AC3 (parity holds cross-language): make webui-test green on the same corpus; atom offsets independently hand-verified.
• AC4 (no scope creep): diff touches exactly the 2 files above; app/, indexer/, non-corpus webui/src untouched.

Review

Independent code-reviewer pass (separate context, not self-review) against the plan's acceptance criteria: 0 findings (CRITICAL/HIGH/MEDIUM/LOW), offsets independently recomputed and confirmed correct, scope confirmed to the 2 files above. APPROVE.

Integration sequencing

Branch off origin/integration/search-hardening-and-ci-integrity, base of this PR is that integration branch (not master) — child of umbrella draft PR #96, which stays open and untouched. No sequencing dependency on sibling issues #38/#75/#80; only shared-file conflict surface is queryModel.corpus.json, a trivial append.