fix: clean up query error listeners

[hsusul]

Fixes #1118

What kind of change does this PR introduce?

Bug fix

What is the current behavior?

Every query adds a one-shot error listener to the shared PostgreSQL pool. Successful queries and query-level rejections do not remove it, so completed queries accumulate stale listeners and eventually trigger MaxListenersExceededWarning.

What is the new behavior?

The query wrapper now settles through shared resolve and reject paths that clear any delayed connection-error timer and remove the exact temporary pool listener. Connection-level pool errors retain their existing next-tick arbitration and still propagate to the caller.

Tests

Added focused mocked-pool coverage for:

• repeated successful queries leaving no residual listeners
• query rejections cleaning up their listener
• active connection-level errors still propagating

Local validation:

• npx vitest run test/db.test.ts — 3 passed
• npm run check — passed
• npm run build — passed
• Prettier check — passed
• git diff --check — passed

Docker-backed integration tests were unavailable because the local Docker daemon is not running.

Additional context

#911 introduced connection-level error propagation but does not clean up listeners after ordinary completion. Open #782 is a broader client abstraction refactor and does not report this listener leak.