fix(auto): rate-limit empty-pool AUTO warnings

[RaviTharuma]

Summary

• virtualFactory log.warns auto/<family> matched no connected models; returning an empty pool on every resolve.
• Observed on 3.8.49: that line ~once per 1–2 minutes continuously when a family (e.g. a virtual auto/<family>) has zero connected models — including when nothing user-facing needs that family (suspicion: health/catalog tick).
• Behavior unchanged (still empty pool). Warn is rate-limited to once per label per 60s.
• Unit test for the debounce helper.

Related Issues

• Empty-pool AUTO log-storm issue filed in this batch
• Related to fix(ops): /healthz (and other HTTP) times out while event loop is busy (k8s probes flap) #10052 only as extra main-thread logging under load

Validation

• Production-code changes include a new or updated automated test in this PR

node --import tsx --test tests/unit/auto-empty-pool-warn-once.test.ts

No secrets or private hostnames.

[RaviTharuma]

Implements / pairs with #10346 (rate-limit the AUTO empty-pool warn). Does not change empty-pool routing.

[diegosouzapw]

Thanks for this — the debounce logic itself is clean and well-tested. I cherry-picked just your virtualFactory.ts commit onto a fresh release/v3.8.50 tip: it applies with zero conflicts, tests/unit/auto-empty-pool-warn-once.test.ts passes, and eslint is clean on the touched files. No concerns with the actual fix.

One thing to fix before merge, though: the PR branch looks like it was cut from main rather than rebased onto release/v3.8.50, so the diff currently also carries 4 commits that aren't part of this fix — the Dependabot nanoid/dompurify bump, the .gitignore _tasks symlink hardening, PR #10026's catalog excludedModels filter, and your own earlier #10055 pricing-memoization fix. All four already live on main but haven't synced to release/v3.8.50 yet, so merging as-is would pull them in ahead of the normal release sync-back and could cause duplicate/conflicting commits later.

Could you reset your branch to the current release/v3.8.50 tip and re-apply just your rate-limit commit on top, then force-push? That should also clear up the "Build" and "Quality Ratchet" CI reds — neither is caused by your change (Build was an infra runner cancellation mid-build; the Quality Ratchet CodeQL alerts are all in unrelated test files that live on the branch already). Once that's done this should be ready to go.

[diegosouzapw]

Babysit summary

• Rebuilt branch onto current base release/v3.8.50 (was 2,280 commits stale). The previous PR head carried leaked unrelated content (deps bumps, catalog/radar changes from other contributors); rebuilt as base + af88f6a55 only. New head: 8bed9d8a9.
• Preserved author + credit: RaviTharuma RaviTharuma@users.noreply.github.com, Co-authored-by: diegosouzapw diegosouza.pw@gmail.com.
• Docs Gates (fast-path) fail = infra, not defect: onnxruntime-node binary download from api.nuget.org timed out (ETIMEDOUT/ENETUNREACH) during npm ci — the docs checks never ran. Re-run of the job passed cleanly.
• Gate: all checks green (Docs Gates, Unit Tests, Vitest, Fast Quality Gates, ESLint warnings, Merge integrity, dast-smoke, semgrep, Build advisory).
• Tests: tests/unit/auto-empty-pool-warn-once.test.ts passes locally; lint clean.
• Ready for human review & merge.