CI on main is red: the e2e job fails when the Chromium download fails #550

What happens

CI on main failed at 6599746 (run 39693276269). The lint-type-test job passed. The e2e job failed at its pnpm exec playwright install --with-deps chromium step with Failed to download Chrome for Testing 147.0.7727.15 (playwright chromium v1217) ... Download failure, code=1. No migration, build or test ran, so nothing in the commit is implicated.

The step has no retry, so one bad download from the browser CDN turns main red. While main is red the pipeline merges nothing, and the pipeline's token can't re-run a job (rerun-failed-jobs returns 403), so main stays red until someone re-runs it by hand or pushes again.

Fix, in outline

• Retry the install. In .github/workflows/ci.yml, wrap the playwright install --with-deps chromium step in a short retry (for example three tries with a pause between them), so a one-off download failure doesn't fail the job.
• Optionally cache ~/.cache/ms-playwright keyed on the installed Playwright version, so most runs skip the download. Keep --with-deps (or playwright install-deps chromium) running either way, since the system packages aren't in that cache.

Watch for

• This touches .github/workflows/, which the pipeline treats as high risk, so the PR waits for the owner before merging.
• Don't retry or loosen the test step itself. Only the browser download is retried.

Done when

• A failed first download no longer fails the e2e job, and a download that fails every time still fails it.
• CI is green on the PR, and on main after it merges.
• pnpm lint && pnpm typecheck && pnpm knip && pnpm test pass.

CI on main failed at 6599746; checked against the job log and .github/workflows/ci.yml at that commit. https://claude.ai/code/session_017Bds4PiJYfNQooN2quaa9Q