test: resume tiered execution and eliminate fixture waits

[backryun]

The test-efficiency work in #12539 stalled between the relocation/tier PRs and an unpublished quick-tier prototype. This continuation adds an opt-in runner derived from the canonical collectors and fixes two measured sources of unnecessary waits without deleting coverage or weakening assertions.

Depends on #12588, then #12606. This PR targets the release branch because the parent branches live on a fork. The release-relative diff therefore includes their relocation changes. Review this continuation's two commits / 12 files here:

Incremental diff against the integration-tier parent

Do not merge this continuation before its parents and the release-base failures are resolved. No auto-merge is enabled.

⚠️ base-red inherited: #12732

The existing base repair candidate is #12333; its changes are not copied into this test PR. This is a non-draft PR so the repository's CI actually runs.

Changes:

• Add npm run test:unit:quick with --list and --concurrency. It excludes the six previously measured heavy groups, preserves per-file process/data isolation, dashboard tsx and serial concurrency, includes new canonical groups automatically, bounds command-line size, and reports failures even when later batches pass. Full CI and coverage commands are unchanged.
• Replace five Batch API polling loops with the existing processing/completion helpers. The success test now uses a deterministic response and requires successful output instead of accepting credential failures. Existing scheduler startup/shutdown coverage remains.
• Replace three Redis test doubles' TCP-chunk matching with a shared RESP-framing fixture. Pipelined/fragmented handshakes no longer strand concurrent-probe and socket-release tests; those assertions remain and each factory test is bounded at 10 seconds.
• Document the recovered rollout, exact commands, measurements, remaining failures, and landing order. The earlier audit found no exact duplicates; no test suites were deleted on similarity alone.

Validation:

• Focused Node regressions: 66 passed.
• Vitest: 465 passed in 51 files.
• Test discovery: 5,536 files / 34 collectors, unchanged frozen orphan debt of 9.
• Targeted ESLint, check:docs-all, diff checks, and normal commit hooks passed. Hooks ran without a shared git stash.
• Batch/processor group: 46 tests passed in 3.05s; the five API cases previously spent about 53s waiting for scheduler ticks. Redis factory group: 5 tests passed in 0.28s after two cases stalled over five minutes.
• The initial quick run selected 3,898/4,946 files and recorded 29,723 passes, 26 failures, and 20 skips. Two failures were the stalled Redis files explicitly terminated during diagnosis; that run is not an uninterrupted success. The later focused runs verify their fixes.
• Six additional failure files were rerun unchanged on the integration-tier parent: the same nine local failures reproduced. The checked-in rollout document lists them so they are not mistaken for quick-runner regressions or waived silently.

Full-suite green status and the complete before/after performance/coverage comparison remain outstanding. The 60/60/60/60 coverage floor is unchanged. Refs #12539 and #12589; this PR does not close the broader overhaul.