chore(hooks): keep mobile analysis out of pre-commit

[wesbillman]

Summary

• run only Dart formatting for mobile changes during pre-commit
• move Flutter static analysis to the path-scoped pre-push graph
• fix case-sensitive Justfile triggers for the Rust and Tauri pre-push gates
• skip whole-tree desktop/web formatting for lockfile-only commits while retaining every lockfile-triggered pre-push check

No test suite was removed or narrowed. Cargo formatting remains workspace-scoped; frontend and mobile source changes still select their existing formatters.

Benchmark

Warm isolated timings on an M2 Max:

• dart format .: 1.92–2.23s
• flutter analyze: 5.58–7.51s
• old forced full pre-commit: 10.16–18.40s
• current forced full pre-commit at f18d9b5: 5.42–7.09s, median 6.29s (45% lower)
• current forced full pre-push at f18d9b5: 2m45s, effectively unchanged from the prior warm 2m41s run
• lockfile-only pre-commit after the follow-up: 0.18–0.19s across three runs

The new mobile analysis lane finishes before the existing mobile test lane, so it added no observed full-push wall time.

Validation

At pushed head 79626be60d89ba34e0fe136f254cc38cf1f1c2b8:

• lefthook validate
• isolated temporary-repository selection test:
  • Justfile selects both Rust and Tauri pre-push gates
  • pnpm-lock.yaml selects neither mutating frontend formatter
  • desktop/web source files still select their formatter
• lockfile-only pre-commit: 0.19s, 0.19s, 0.18s
• pre-push hook passed on the exact pushed head

Earlier full-cycle validation at f18d9b5802f11543e8afbef3cd54c5928817e32b:

• forced full pre-commit: 5.42s, 6.29s, 7.09s
• forced full pre-push: 2m45s; all lanes passed
• mobile tests: 1,465 passed

[jedwards27]

:bot: Jude’s code review agent

Verdict: APPROVE

Reviewed: d2cfd377e27dab8fdef0236dd8e92c89efbae829..f18d9b5802f11543e8afbef3cd54c5928817e32b

Risk: Low — one-file local hook-graph change; no runtime, protocol, persistence, identity, packaging, or release contract changes.

No material findings. I traced the formatter/analyzer/test recipes and the Lefthook selection, failure, and parallelism semantics:

• lefthook.yml:39-42 now runs formatter-only just mobile-fmt for mobile/**, retaining stage_fixed.
• lefthook.yml:82-87 runs mobile-check and the existing mobile-test in parallel for the same mobile path shape. mobile-check still performs non-mutating format validation plus flutter analyze; mobile-test still runs the Flutter suite.
• An isolated Lefthook 2.1.3 semantics probe using these command blocks started both pre-push jobs concurrently, waited for both, propagated a failing mobile-check through the aggregate exit status, and skipped mobile formatting for a non-mobile input.
• A mutation probe added an undefined Dart function: pre-commit formatted and passed as intended, while pre-push failed with the exact file/line and undefined_function diagnostic. Analysis is deferred, not discarded.

Exact-head validation:

• git diff --check d2cfd377e27dab8fdef0236dd8e92c89efbae829...f18d9b5802f11543e8afbef3cd54c5928817e32b — pass
• lefthook validate under pinned Lefthook 2.1.3 — pass
• representative mobile pre-commit — formatter only, pass
• representative non-mobile pre-commit — mobile command skipped, pass
• just mobile-check — pass, no analyzer issues
• just mobile-test — pass, 1,465 tests
• representative mobile pre-push — both jobs selected and pass; tests remained the wall-clock critical path

The diff matches the documented hook contract in AGENTS.md:119-130 and introduces no unrelated scope or conflict with VISION.md / TESTING.md.

Residual risk: GitHub path filtering skips Mobile because this PR changes only lefthook.yml, so CI does not exercise the changed hook shape. Exact-head local selection, mutation, analyzer, and full mobile-suite evidence covers that gap. The aggregate Desktop check is red because smoke shards 1 and 3 were cancelled; successful sibling smoke shards and the one-file hook-only scope make that unrelated to this verdict.