perf(route-cache): V5 passive SQL settlement and event-loop root-cause diagnostics

[Boym323]

V5.1 – Read-only route-cache DB latency attribution

• Measures actual settlement of every initiated DB-call promise, including failed and late-settling operations, without changing its behavior.
• Measures timer-callback overshoot beyond the existing 750 ms deadline; reports process event-loop utilization during caller timeouts.
• Distinguishes wall-time DB call (including pool/network) from NOT MEASURED server-side SQL and pool wait.
• Includes operator-run read-only PostgreSQL diagnostic SQL (aggregate activity, indexes and planner-only EXPLAIN).

V5.2 – Event-loop correlation and bounded soak projection

• Adds bounded process-local ELU ring; extends read-only T5.7 sampler and report.
• Explains limitations of overlapping CPU attribution, measurement scopes, and privacy constraints.
• Adds English and Czech documentation.

Checks

• CI PASS: build, docs:check, lint, typecheck, Vitest, migration integrity
• CodeQL PASS
• Added focused regression tests for timeout lifecycle and bounded telemetry.

Safety and release

No change to existing 750 ms caller timeout, circuit breaker/backoff, SQL operations, cache semantics, configuration, or DB schema/data. No manual production mutations; standard validated main CI/CD release is the only deployment path. Independent SQL script must be run only by operator with least privilege. Follow with a short, passive 1–2 hour V5 production measurement after deploy and version identity validation.