Follow-ups: E2E CI tightening (retries, scoping, shard reuse, seeding, dedup)

[chiptus]

Parent

#217

What to build

A grab-bag of deferred follow-ups from the E2E speed-up work (#217–#223, shipped in #224/#225), each independently actionable:

1. Drop smoke suite retries from 2 to 0 once the smoke workflow (e2e-smoke.yml) has proven stable for a reasonable stretch post the wait-time fixes in Fix real-time waits in E2E specs #218. Was deliberately left at 2 initially to avoid immediately blocking PRs on flakes (see Speed up E2E tests: smoke suite for PRs, nightly full suite, fake clocks #217 user story 11).
2. Changed-file-based test scoping for the smoke suite — map changed source files to related specs so the smoke suite can run more than the fixed 5-spec critical path when relevant, instead of the same fixed set on every PR (Speed up E2E tests: smoke suite for PRs, nightly full suite, fake clocks #217 user story 12 / Out of Scope).
3. Build-artifact reuse across nightly matrix shards — build once and share across the 4 shards × 3 browsers instead of rebuilding per-shard, to cut nightly full-suite runtime (Speed up E2E tests: smoke suite for PRs, nightly full suite, fake clocks #217 user story 13 / Out of Scope).
4. Lighter/faster Supabase seeding to replace supabase db reset in CI, investigated as a possible speed win (Speed up E2E tests: smoke suite for PRs, nightly full suite, fake clocks #217 user story 13 / Out of Scope).
5. De-duplicate shared setup between e2e-smoke.yml and e2e-tests.yml — flagged in code review of ci(e2e): split PR smoke suite from nightly full suite #224: the two workflows largely duplicate their setup steps (checkout, pnpm/node setup, install, Playwright-version lookup, browser cache, conditional install/install-deps, Supabase start/reset, build). Extract into a reusable composite action or workflow_call so a future change to Node/pnpm/Playwright caching doesn't need editing in two places.

Acceptance criteria

• Smoke suite retries dropped to 0 (or a documented decision to keep 2, if instability is found)
• A changed-file → spec mapping (or explicit decision not to pursue it) is in place for the smoke suite
• Nightly workflow builds once and reuses the artifact across shards, or a documented reason not to
• Supabase seeding in CI is faster than a full db reset, or a documented reason not to change it
• Shared setup steps between the two E2E workflows are deduplicated into one reusable unit

Blocked by

• None — can start immediately (each item is independently actionable; no need to do them together)