Retry E2E tests once in CI and report flaky results

[isaacroldan]

WHY are these changes introduced?

The E2E suite runs with retries: 0, so one flaky interaction fails the entire shard. Retrying alone would just hide the flake, so this pairs the retry with reporting.

WHAT is this pull request doing?

• retries: isCI ? 1 : 0 — a test that fails once and passes on retry keeps the shard green. Local runs keep zero retries so flake fails fast during development.
• maxFailures: 3 → 5 in CI — first attempts of flaky tests count toward the early-stop limit, so the old value could kill a shard before retries got to prove tests flaky rather than broken.
• JSON reporter in CI + scripts/report-flaky.js — every failed-then-passed test is printed as a grep-able FLAKY: log line and listed in the GitHub job summary, so the nightly failure counter (and humans) can track flake instead of losing it to green shards.

Interaction with the cli-kit throttle retry (#8317)

Layered deliberately: the CLI-level retry absorbs short throttle bursts inside a command; the Playwright retry covers everything else. Worst-case per-test wall clock grows — if shards start brushing the 20-minute job timeout, that's the knob to revisit.

How to test your changes?

cd packages/e2e
mkdir -p test-results && echo '{"suites":[{"title":"x.spec.ts","suites":[{"title":"S","specs":[{"title":"t","file":"x.spec.ts","tests":[{"status":"flaky"}]}]}]}]}' > test-results/results.json
node scripts/report-flaky.js   # prints FLAKY: x.spec.ts › S › t

Measuring impact

• n/a — CI/test-infra only, no user-facing impact

🤖 Generated with Claude Code