fix(tests): give two default-vitest-timeout cases real margin - #9368

Fixes #9311

What

Two explicit timeouts, exactly the scope the card's triage comment fenced:

1. packages/types/src/node.test.ts:178 — "falls back to the importing package's own resolution when the host does not declare" gets an explicit 30s testTimeout. It is the only case in its describe block that performs a real dynamic import() of @objectstack/spec (a multi-megabyte package); every sibling resolves a small on-disk fixture or fails fast, all under 10ms — checked directly, not assumed. Measured on this box: ~0.9-1.1s unloaded across 3 isolated runs. The card's own isolation runs already showed it failing at 5061ms against the 5000ms default under nothing heavier than turbo run test --concurrency=2, so 30s (matching the repo's #3662 precedent for subprocess/real-load cases) gives real headroom without masking a genuine hang.

2. packages/qa/dogfood/test/semantic-roles.dogfood.test.ts:52 — the beforeAll that boots the full showcase stack (ObjectQL + ~45 plugins) through @objectstack/verify's bootStack gets an explicit 180s hookTimeout. This file's own isolated run here measured 18.3s (vitest Duration) / 19.5s wall clock for the whole file (boot dominates; the 5 tests themselves run in ~3.5s) — in the same range as the card's own ~27s unloaded measurement. The previously observed failure was 10027ms against the 10000ms default, essentially zero margin even idle. 180_000ms follows this package's own existing house pattern for the identical bootStack(showcaseStack, …) call — admin-identity-audit-trail.dogfood.test.ts's beforeAll(…, 180_000) — rather than inventing a new number.

No sibling dogfood file carries a package-level hookTimeout override (vitest.config.ts sets none), so the per-hook precedent above is the house pattern being followed.

Why this and not turbo.json concurrency

Both cases pass in isolation and the repo's full suite is confirmed green at --concurrency=2 --continue (comment on #9311). The margin was never sized for the work these two cases actually do (a real package import; a 45-plugin stack boot) — that is the "budget never sized for the work" case the dispatch explicitly distinguished from "hides work that got slower." Neither timeout changes behaviour; both suites already pass today. turbo.json's default concurrency is out of scope per the card's own filing (a maintainer-level default) and is not touched here.

Verification

• packages/types: tsc --noEmit clean; vitest run — 348/348 tests pass, matching the card's own suite-size measurement.
• packages/qa/dogfood: tsc --noEmit clean; the changed file alone — vitest run test/semantic-roles.dogfood.test.ts — 5/5 pass, run 3× for timing consistency (Duration 18.26s / 19.5s wall clock).
• Static gates run clean at head dc151fd0 (all exit 0): check:test-source-alias, check:type-source-resolution, check:query-options-erasure, check:engine-double-contract, check:where-matcher, check:cross-package-test-inputs, check:slot-lookup, check:nul-bytes, check:type-check-coverage, and the packages/spec liveness family (check:empty-state, check:liveness, check:strictness-ledger, check:variant-docs), plus scripts/docs-audit/check-affected-docs.mjs.
• check:type-check-debt (the re-measure half) could not complete locally — it needs the full 78-package workspace closure built (pnpm exec turbo run build --filter='./packages/*' --filter='./packages/*/*') and this worktree hit a missing @objectstack/service-knowledge dist partway through an unrelated package's dependency chain. Neither @objectstack/types nor @objectstack/dogfood appears in the DEBT/TEST_DEBT ledgers in scripts/check-type-check-coverage.mjs (grepped directly), so this diff — test-file-only, no new imports, no new test files — cannot move that ratchet; CI's own build will run it in a clean environment.
• Did not run the full @objectstack/dogfood package suite (~780 tests across ~100 stack-booting files) locally — a real attempt ran past 8+ minutes without finishing under --maxWorkers=2 on this shared, multi-agent-contended box, and continuing would hold the container-wide heavy-verification lock for an extended period at every other parallel agent's expense for a change with zero behavioural surface. The specific changed file was verified directly (above); CI runs the full suite in its own environment.

Generated by Claude Code