ci: web/frontend has no build gate and zero tests — three UI bugs shipped past CI in one PR

[padak]

PR #658's review round demonstrated the gap concretely: three real frontend bugs (a payload field mismatch hiding the re-run button, an optimistic-update leak, a missing branch_id on re-run) were caught only by manual review and a live browser pass — none by CI, because no workflow builds web/frontend at all.

Two decisions to make:

1. Build gate (cheap, high value): add a CI job running cd web/frontend && npm ci && npx tsc --noEmit && npm run build on PRs touching web/frontend/**. This alone would catch type-level payload mismatches like the configId vs config bug (job list: Config column has always rendered empty — renderer reads an invented 'configId' key #659 is the CLI-side twin).
2. Tests (bigger call): vitest currently reports "No test files found" — there are zero test files under web/frontend/src. Decide whether the frontend gets a test harness (component tests for payload-shape contracts against server/routers/* models would have caught all three bugs), or whether the build gate + review is the accepted bar.

Context: web/frontend is ~15k LOC and growing (PR #658 added a Tokens page, trash/restore, command palette...). The Python side has 6000+ tests; the UI has none.

[padak]

Decision 1 (build gate) shipped in #661 (commit 1ee9ab1): .github/workflows/frontend.yml runs npm ci + tsc --noEmit + vite build, path-filtered on web/**. That closes the specific gap this issue was opened for — the configId vs config payload mismatch is exactly the class of bug tsc --noEmit now catches before merge.

Decision 2 (a frontend test harness) is deliberately not part of that PR: vitest with zero test files exits 1, so wiring it in as a required step would have made the gate fail on every PR. Closing this issue as the build gate is done; if we want component tests for payload-shape contracts against server/routers/* models, that is worth its own issue with its own scope rather than lingering here as an undecided second half.