test: remove quick-tier batch barriers and fixture waits

[backryun]

The quick tier previously stopped at nine sequential command-line batches. This continuation gives each phase one scheduler, preserves process isolation, and removes measured timer and host-environment dependencies from fixtures. The earlier local comparison of the same 3,903 selected files at concurrency 8 improved from 391.26s to 310.36s (20.7%); that benchmark predates this base refresh and has not been repeated.

Depends on #12588, then #12606. Restacked onto release/v3.8.51 at d6f3150. The release-relative diff includes both parents because their branches live on a fork. Review this continuation's four commits / 24 files in the incremental parent comparison.

Changes:

• Derive the opt-in quick tier from canonical collectors, retaining the six measured exclusions, dashboard loader, serial concurrency, and file isolation. Pass paths through stdin to remove command-line limits and batch barriers.
• Clear inherited application DB overrides before the per-file isolation preload. Record JSONL timings, phase results, and worker exits; drain reporter output and preserve failure, skip, TODO, and leaked-handle behavior.
• Replace timer waits in Batch, compression, provider/combo recovery, and JobRegistry fixtures with existing helpers and controlled clocks. Repair Redis command framing and calendar, CLI, janitor, binary-path, and mocked Adobe fixtures without dropping assertions.
• Record the base refresh, CI shard correction in the parent, and current validation in docs/architecture/TEST_EXECUTION_TIERS.md.

Current validation on Node 24.20.0:

• All 14 quick runner/worker regressions passed in the full Node run. The plan selects 3,927 of 4,985 canonical Node files.
• Full Node main: 37,675 passed, 34 failed, 23 skipped. Every failure name also appeared in the release-base run (37,698 passed, 43 failed, 23 skipped). These are separate runs with different partitions and scheduling, not a controlled speed or assertion-count comparison.
• Dashboard separately: 163 passed, one failure with the same redaction assertion as the base. Serial: 23 passed.
• Vitest: 465 passed in 51 files after initializing the isolated temporary DB. The initial attempt had a concurrent migration collision and skipped five tests in one failed suite.
• Parent relocation/conflict tests passed 366 and 37 tests; partition/serial guards passed 15. Discovery passed with 5,576 files, 34 collectors, and nine frozen orphans.
• Integration shard execution found shared-harness import drift, now fixed in Test tiers: split slow e2e and upstream-gated live out of default integration glob #12606. The remaining six combo assertion failures reproduced on the release base. Live collection passed one mocked test and skipped 17 opt-in tests. Resilience HTTP failed and hung, requiring termination.
• Runner-API and pre-commit checks passed. Strict mutation registration retains two base omissions. check:docs-all retains six base count drifts (171 migrations versus 169; 20 routing strategies versus 19).
• Full coverage was not rerun; the 60/60/60/60 floor is unchanged. CI must be checked on the newly pushed heads.

⚠️ base-red inherited: #12732. The separate production repair candidate remains #12333. Full Node, integration, mutation, and documentation gates are still red; this does not establish a green combined release base. Auto-merge is not enabled.

Refs #12539 and #12589. Landing order remains #12588 → #12606 → this continuation.

[backryun]

Superseded by #13163, which rebuilds the useful deterministic fixture changes as one focused commit on the current release base. The replacement preserves existing test paths and CI collection. This branch is retained for reference.