Playwright end-to-end gate: composite and e2e-playwright reusable workflow (#75)

[ms280690]

Closes #75. Stacked on #147 (issue-72-test-node) for the Node setup pattern (setup-node, pnpm/action-setup, lockfile detection); retarget once that merges.

What

• .github/actions/playwright: Node version and package manager from the lockfile (same detection as node-test), setup-node with the package-manager cache, install, read the installed @playwright/test version, actions/cache on ~/.cache/ms-playwright keyed on that version and the browser list, playwright install --with-deps (only install-deps on a cache hit), then playwright test --reporter=github,junit,blob with --project and --shard i/n when given and BASE_URL exported from target-url. summary.py (stdlib) turns the JUnit XML into a per-file table and a failure list in the step summary and the tests-total, tests-failed, tests-skipped outputs; annotations come from the github reporter (file and line). HTML report built from the blob with merge-reports. Artefacts: JUnit, blob and HTML always (<name>-results[-shard]), test-results/ with traces, videos and screenshots on failure (<name>-traces[-shard]). Final gate step fails on a non-zero playwright test exit.
• e2e-playwright.yml: inputs from the issue plus browsers, extra-args, runtime-version, install-command, actions-ref. shards job builds the matrix, test runs one composite per shard (fail-fast: false), report (only when shards > 1) downloads the shard blobs and runs npx @playwright/test@<same version> merge-reports --reporter html into playwright-report.
• Fixture project tests/e2e-playwright/fixtures/project: @playwright/test 1.64.0 pinned with a lockfile, server.js as the webServer, two passing specs and a failing spec enabled by E2E_FIXTURE_FAIL; playwright.config.js is the documented config contract (BASE_URL, trace: retain-on-failure).
• CI E2E Playwright job: summary.py fixtures, the composite on the fixture (3 tests, 0 failed), then with E2E_FIXTURE_FAIL=1 expecting failure with tests-failed=1.
• README rows and "E2E (Playwright)" section. No new third-party actions beyond those feat(#72): Node test gate — vitest / jest reusable workflow #147 already approved.

Tests

tests/e2e-playwright/run.sh with recorded Playwright JUnit (pass with a skipped test, one failure): counts, outputs, headings with and without a shard, suite rows, failure message, no duplicate annotations, exit codes. Verified locally with Playwright 1.64.0: passing run, failing run (annotation ::error file=tests/failing.spec.js,...,line=7), --shard 1/2 blob and merge-reports --reporter html.

Deferred

• Run against one Sparkgeo frontend repository after merge.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NPH9145e4gFo6bmXvMgBoF