fix(tests): resolve remaining flaky tests after Vitest migration

[vseehausen]

Summary

Fixes three tests that kept flaking after #3139. All three were visible in Datadog CI Pipeline Health but missed by our own flaky-test report, whose log parser is Jest-specific and went blind once the suite moved to Vitest (#3136).

• name.builder — nameBuilder() sanitized (trimmed) the faker name and then truncated to NAME_MAX_LENGTH; the cut could land right after a space and re-introduce trailing whitespace, so sanitizeName(name) !== name. Now re-sanitizeName() after truncation so the result is a fixed point.
• bridge-name.entity.spec — the "unknown bridge name" case used faker.word.noun(), which can itself be a bridge name (omni, polygon, across, gnosis, …), making safeParse succeed and the assertion flip to expected false to deeply equal …. Now resamples until the word is genuinely not a bridge name.
• Integration specs (61 files) — afterEach/afterAll called await app.close() unguarded, so any beforeEach failure left app undefined and every test in the file reported Cannot read properties of undefined (reading 'close') instead of the real error (the "27 tests fail at teardown" cascade). Guarded with await app?.close().

Testing

• yarn test:unit for name.builder + bridge-name.entity → pass
• biome check clean; tsc introduces no new errors

Related

• WA-1753
• Follows chore(tests): reduce CI flakiness (clearMocks + seeded faker) #3139 (clearMocks + seeded faker)

[PooyaRaki]

@claude review

[claude]

Claude finished @PooyaRaki's task in 1m 33s —— View job

---

Claude PR Review (re-review)

• Check follow-up commit ce81716 (dropped if (app) guards)
• Verify name.builder.ts fix
• Verify bridge-name.entity.spec.ts fix
• Sample integration specs for consistency
• Post final review

---

Follow-up looks good — LGTM. No further blockers from me.

Verified

• ce81716 (if (app) guards dropped) — grep for if (app) across **/*.integration.spec.ts returns zero matches; the two touched files (feature-flag.service.integration.spec.ts:23, messages.controller.integration.spec.ts:49) now use the plain await app?.close() pattern used everywhere else in the PR. Consistent across all 63 files.
• name.builder.ts — re-sanitizeName() after codepoint-aware truncation is correct; [...padded].slice(0, NAME_MAX_LENGTH).join('') handles surrogate pairs safely, and the second sanitize normalizes any whitespace re-introduced at the cut boundary.
• bridge-name.entity.spec.ts — noted your reasoning for keeping the unbounded while loop (avoiding a deterministic fallback that undermines the "random non-bridge noun" intent). Fair trade-off given faker.word.noun() draws from a very large vocab vs. ~20 bridge names — non-termination is astronomically unlikely with a seeded faker.

Nits (non-blocking, informational)

None new. Previously-raised #3 (theoretical NAME_MIN_LENGTH violation after trailing-whitespace trim) remains not reachable in practice with the current SAMPLE_NAMES + faker.person.fullName() inputs.

Ship it.
• branch fix/wa-1753-remaining-flaky-tests