feat(resilience): expose live combo queue telemetry

[xiaoyaner0201]

Summary

Adds a live queue observability panel to the System Health dashboard so operators can see where concurrency is accumulating without changing routing behavior.

Closes #15085
Related to #7778, #3872, and #8884

Changes

• Expose the existing round-robin combo semaphore snapshot through the authenticated GET /api/admin/concurrency endpoint as comboQueues.
• Add optional namespace filtering to rateLimitSemaphore.getStats() so the API reports only actual combo: gates for the combo view.
• Add a dashboard panel that refreshes every 15 seconds and keeps combo-target queues separate from account semaphore queues.
• Show running, queued, configured limit, cooldown/blocked state, snapshot timestamp, and load errors.
• Add focused tests covering namespace filtering, API scope/auth wiring, and dashboard scope/error/timestamp presentation.

Validation

• ✅ node --import tsx/esm --test tests/unit/queue-observability.test.ts — 3/3 pass
• ✅ npm run typecheck:core
• ✅ npm run lint -- --pass-on-unpruned-suppressions
• ✅ Targeted ESLint on all changed files
• ✅ npm run build
• ✅ git diff --check
• ⚠️ npm run lint without the existing --pass-on-unpruned-suppressions override exits because the repository currently reports unused suppression entries; the same full lint passes with the repository-supported override.
• ⚠️ npm run test:scoped timed out while rebuilding the repository-wide import-impact map; the focused changed-area test passes independently.

Scope / non-goals

This PR is observability-only. It does not change queue limits, routing, fallback, account selection, historical retention, per-request wait attribution, or cross-process aggregation. Queue depth is an in-memory point-in-time snapshot and resets with the process, consistent with the existing semaphore implementation.