Pin retries: 0 for playwright-wrapped flake/failing tests

Agreed in discussion 2026-09-04. Playwright has no per-test retry option (vitest does — the wrappers already pass { retry: 0 } there), so the playwright branch of createFlake/createFailing registers inside an anonymous test.describe scope carrying test.describe.configure({ retries: 0 }).

Why#

The green paths were already structurally unretryable (both end in a throw that satisfies test.fail, and playwright only retries unexpected outcomes). The leak is the RED path: "passed unexpectedly" IS unexpected, so playwright retried it — double-recording ❌ and occasionally rescuing the run into a green that distorts the record stream. With the pin, both runners have identical semantics: a wrapped test never retries, period.

An anonymous describe keeps identity intact: it contributes an empty title segment that titlePath().filter(Boolean) drops, so the leaf title — the dashboard row key — and reported fullName are unchanged.

Checklist#

• createFlake: playwright branch registers via anonymous describe + configure({ retries: 0 }).
• createFailing: same.
• Faithful fakes, not production fallbacks (Misha's call): the unit-test fake playwright objects grow describe/describe.configure (invoking the callback synchronously), and the tests assert the configure call — no "describe" in test hedging in the wrappers.

Notes#

• #2575 (open) restructures failing-test.ts's registration branch around "setTimeout" in test; this PR keeps the existing "fails" in test split and only wraps the else branch, so whichever lands second does a small fix-up.