[krazymir]

What

The Jenkins e2e now runs as one Playwright run with E2E_WORKERS workers (default 4) instead
of E2E_SHARDS separate Playwright containers. Test and CI only, so there's no version bump and no
changelog entry.

Why

The shards did by hand what Playwright workers already do: N containers with one worker each, plus a
greedy balancer in playwright.config.ts that kept the specs changing platform-wide settings
together in shard 1. Playwright 1.52+ can cap the workers of a single project, so that rule is now
config:

• platform-state: the 13 specs that change platform-wide settings (AI provider, banner,
  settings, webhooks, featured, gamification, hype, public form, surveys), workers: 1. They still
  run one after another. The project is listed first, so whenever its one worker frees up it takes
  the next of those files at once; Playwright 1.64's dispatcher picks the first queued file whose
  project is under its cap.
• isolated: everything else, given to whichever worker is free.

Expected speed-up

The 29 platform-state tests (of 112) must run back to back, and that sets the floor on the wall
time. Today's default of 2 shards runs about 55 tests end to end per shard. 4 workers bring that
down to the 29-test chain, roughly 2× faster. Beyond about 4 workers the run gets no shorter;
each test just slows as more browsers share the one dev server. Hence the default of 4. These are
estimates by test count, not measured durations.

Changes

• e2e/playwright.config.ts:
  • E2E_SHARD/E2E_SHARD_TOTAL and the shard balancer are gone. E2E_WORKERS takes 1–16
    (default 4); any other value fails fast.
  • Whole files are dealt out to workers (fullyParallel stays off), so two tests of one file
    never overlap, exactly as with shards. Splitting files would gain nothing: the platform-state
    chain is far longer than any one file (largest: 7 tests).
  • One HTML report instead of N.
  • maxFailures: 10 now counts across the whole run; before, the limit applied to each shard
    separately.
• Jenkinsfile:
  • The E2E_SHARDS parameter becomes E2E_WORKERS (default 4) and applies to the quick run too.
  • One Playwright container, one exit file, no [shard i] log prefixes.
  • The dev-server watchdog, host stats and post-mortem are unchanged; the watchdog now stops one
    container instead of N.
  • The first build after this lands falls back to 4 until Jenkins picks up the new parameter.
• e2e/helpers/stamp.ts: uniqueStamp() replaces Date.now().toString(36) at all 107 call
  sites in 46 specs. It adds a letter for the worker (from TEST_PARALLEL_INDEX), so two workers
  starting a test in the same millisecond can't create identically named fixtures. Shards already
  carried that risk. The longest stamped namespace slug is 21 of the 40 characters allowed.
• Comments in scripts/e2e-specs.mjs and public-submissions.spec.ts no longer mention shards.

Checked

• playwright test --list: 29 tests in platform-state, 83 in isolated, every spec in exactly
  one; E2E_WORKERS=0 is rejected.
• pnpm typecheck passes; web unit tests: 570/570 pass.
• The e2e stage's shell script passes bash -n.
• The changed e2e files type-check under a temporary tsconfig. pnpm typecheck doesn't cover
  e2e/, and that check also found one error that predates this PR (sign-out.spec.ts:42),
  left out of scope.
• Not run: the e2e suite itself and the Jenkins pipeline. The first Jenkins build of this
  branch is the real test.

For the reviewer

• Watch the first full run's wall time and e2e-host-stats.log: 4 browsers against one dev server
  is more concurrent load than today's 2.
• A new spec that changes platform-wide state must be added to PLATFORM_STATE_SPECS, as before.