ci(#564): run the Playwright suite on pull requests (Chromium, path-gated)

[BorisTyshkevich]

What & why

Closes #564. The Playwright suite was gated to tags, schedule and manual dispatch, so the e2e job reported skipped on every pull request and every push to main — while the overall run reported success. A check that never runs is not a check.

The cost was concrete, not hypothetical. Today's bug/low batch (#554–#558, #561–#563) all merged on green CI; two of those PRs changed and added e2e specs that CI never executed. Dispatching the workflow by hand against main gave the first real signal in that range — 6 failed / 568 passed, the same two tile-open-workbench specs failing on chromium, firefox and webkit. A second dispatch at f68861c produced the identical 6 failures, so they pre-date the batch. Filed as #565.

The change

Pull requests now run Chromium only, path-gated on anything that can move rendered geometry:

src/**            ← styles.css lives here; the single biggest reason for this filter
schemas/**  tests/e2e/**  playwright.config.js  build/**
package.json  package-lock.json  .github/workflows/**

Tags, nightly and manual dispatch keep the full three-engine matrix. That split is deliberate: both #565 failures reproduce on Chromium, so one engine would have caught them and is cheap enough to sit on every relevant PR — but engine-specific breakage is real in this repo (WebKit focus and IndexedDB behaviour especially), and that is what a release has to clear.

gate needed no change: it already fails on any failure among its needs and tolerates a legitimately skipped job.

Quarantining #565 so the gate is useful immediately

Turning on a gate against a red baseline just blocks everything, so #565's two specs are marked test.fail().

Deliberately not test.skip/test.fixme: those stop exercising the path, and nobody would notice when the bug is fixed. With test.fail() the specs keep running, and the day the underlying defect is repaired they report "expected to fail, but passed" — which is the signal to delete the quarantine. Both call sites say so, and say explicitly that the assertions encode #535's intended widen semantics and must not be rewritten to match the broken output.

Verification

Locally, npx playwright test tests/e2e/tile-open-workbench.spec.js --project=chromium → 21 passed, exit 0 (the two quarantined specs register as expected failures rather than errors).

The real proof is this PR itself: it touches .github/workflows/** and tests/e2e/**, so the path filter matches and the e2e job should run here instead of reporting skipped — which is exactly the behaviour being added. Please confirm that job appears and is green before merging.

Checklist

• npm test passes (the per-file coverage gate is non-negotiable) — unaffected; no src/ or unit-test change
• Tests added/updated in the same change as the code
• npm run build succeeds (single-file dist/sql.html) — unaffected
• Layers kept honest: pure logic in src/core/, network in src/net/ (injected fetch), DOM in src/ui/
• No new runtime dependency
• README / CHANGELOG.md ([Unreleased]) updated if behavior or the deployed surface changed — n/a, CI-only, no shipped-artifact change
• Reconciled affected tracked work (roadmap Roadmap to 1.0.0 #68, the issue body, ADR/CHANGELOG) if this change reshaped it — Two tile-widen e2e specs are stale against the grafana-grid@2 placement shape #565 records the quarantine and its exit condition

🤖 Generated with Claude Code

https://claude.ai/code/session_01GiubaoqEuBzAyo5C4P8Vqr