ci: path-gate static ratchet scans in lint

[JustinTong0323]

Motivation

The lint workflow runs pre-commit --all-files on every PR, so the check-static-ratchets hook (pass_filenames: false) rescans the whole python/sglang tree even for PRs that change only docs or CI config. This is PR 2 of the lint follow-up plan, stacked on #34864 (this branch contains its commit; the diff shrinks to lint.yml alone once that merges).

What changed

• Extended the existing dorny/paths-filter@v3 step with a ratchet filter covering python/sglang/**, the ratchet aggregator and every checker module it imports, its test files, and both gate definitions (.pre-commit-config.yaml, .github/workflows/lint.yml).
• The pre-commit step builds SKIP conditionally: a PR touching none of those paths runs with SKIP=no-commit-to-branch,check-static-ratchets, and the decision is echoed in the log.
• Fail open: only a literal 'false' output skips the check; unclassified events (push, classification failure) keep running it — same semantics as the rust-workspace gate.
• The local check-static-ratchets hook definition is untouched, so local pre-commit behavior is unchanged.

Verification

• Positive path: this PR itself changes .github/workflows/lint.yml, which is in the filter — the lint run on this PR must show SKIP=no-commit-to-branch and the ratchet hook executing. Confirmed: run 31819219123 (10m05s wall) logged ratchet changes: true; SKIP=no-commit-to-branch with validate static runtime ratchets and black-jupyter both Passed.
• Negative path: will link a docs-only run showing SKIP=no-commit-to-branch,check-static-ratchets and the hook banner as Skipped (CI duration reported as evidence, not a threshold).
• Local: PyYAML structure assertions on the new filter and step; shell simulation of the SKIP logic for true/false/empty outputs; pre-commit run --files on the changed file passes.

---

CI States

Latest PR Test (Base): ✅ Run #31819219562
Latest PR Test (Extra): ❌ Run #31819219544