ci: run Playwright e2e suite on every PR

[zphillips]

Field	Value
Change type	♻️ same behavior (chore / ci)
Risk severity	S4 low

Summary

Add a CircleCI e2e job that runs the Playwright regression suite on every branch / PR, so breakage surfaces automatically instead of relying on someone remembering to run yarn test:e2e locally. Follow-up to #34.

Decisions & alternatives

• Approach & why: new e2e job on cimg/node:20-browsers, no branch filter so it runs on all PRs. Kept separate from the existing deploy_testER job, which stays on Node 14 — Playwright needs Node ≥18, so this job rents its own newer image; the two never share a machine.
• Alternatives considered & why rejected:
  • Reuse the deploy job's Node 14 image — rejected, Playwright requires Node ≥18.
  • GitHub Actions instead of CircleCI — open question (see note below). Kept this on CircleCI to match the existing pipeline, but happy to switch if the team prefers?
• Trade-offs / what could go wrong: the webpack-4 + --openssl-legacy-provider dev server booting under Node 20 in CI is the main unknown; may need a small tweak once it runs live.

Links

• Design doc / two-pager: N/A
• Issues / related PRs: Depends on [1.1] test: add Playwright e2e regression net for current app #34 (the e2e suite + test:e2e script). Merge [1.1] test: add Playwright e2e regression net for current app #34 first — this job has nothing to run until those files are on develop.

Tests — how do we know it's safe? (check all used)

• Unit
• Integration
• Contract
• Synthetic / e2e — the suite this job runs passes locally (7/7 green on Node 20)
• Manual — config.yml YAML structure validated locally
• None

⚠️ CI-side verification is still pending: CircleCI won't run on PRs from a fork until "Build forked pull requests" is enabled (admin setting), so this job's first real run happens after that toggle.

Incident detection — if this breaks in prod, how will we know?

CI-only change, no runtime / prod impact. A misconfigured job shows red on the PR in CircleCI — that's the signal.

Incident impact analysis — can we measure the blast radius?

Blast radius is limited to CI. Until this check is marked required in branch protection, a failing or broken job blocks nothing and has zero user impact.

Rollback & mitigation — if it goes wrong, how do we fix it fast?

• Safe to plain-revert? Yes — single-file change to .circleci/config.yml; reverting removes the job cleanly.
• Rollout: direct.
• If revert isn't enough: N/A.

---

To make this an actual merge gate (admin only)

Two settings I can't reach as an outside contributor:

1. CircleCI → Project Settings → Advanced → enable "Build forked pull requests" (so it runs on fork PRs at all).
2. GitHub → Settings → Branches → branch protection for develop → "Require status checks to pass before merging" → select the e2e check. The check only appears in that list after it has run once, so do updated tracks #1 first.