Async executor model for Data APIs

We reworked the threading model in the Data query service so long-running queries no longer block the request-handling layer.

1. All metrics, breakdown, errors and views endpoints now run on a dedicated executor pool.
2. Health checks were moved onto the same model so liveness probes stay responsive even when downstream systems are slow.
3. The synchronous request filter was replaced with a non-blocking equivalent.
4. Connection-pool sizing was tuned and a per-query timeout was added so a single slow query can’t impact other tenants.

Metadata fix: Types column shows [“improvement”, “api”] — “api” is misplaced. Change to [“improvement”].