chore(database): remove the oven-sh/bun#42804 broken-pool detector, fixed in Bun 1.4.4 #2979

What

Bun 1.4.4 fixes oven-sh/bun#42804 (fixed by oven-sh/bun#45304, which ports oven-sh/bun#40913: onconnect never comes after onclose). Stacks' detector for pools broken by that bug is now dead code, and its own tripwire tests say to remove it.

Evidence

On Bun 1.4.4 (the version #2975 moves Stacks to), the two tripwire tests fail exactly as designed:

• core/database/tests/broken-pool.test.ts
• core/router/tests/health-broken-pool.test.ts

Both get Postgres's own error, terminating connection due to administrator command, instead of Bun's broken-slot connection must be a PostgresSQLConnection. On Bun 1.4.2 (control) they still reproduce the broken pool, 16/16 passing.

This Bun no longer leaves the pool broken the way oven-sh/bun#42804 describes. ... If this Bun carries the fix, remove the detector: core/database/src/broken-pool.ts, everything that feeds or reads it, and the tests and fixtures built on this server.

Scope

• core/database/src/broken-pool.ts and its exports (index.ts, runtime.ts)
• utils.ts: the persistent onQueryError hook on PostgreSQL/MySQL, recordPrimaryDatabaseQueryError, the replica builders' hooks
• core/router/src/health.ts: the broken-pool 503
• comments that cite the bug: ensure-database.ts, config/database.ts, docs/guide/query-monitoring.md, the stacks-database skill, the native catalog
• tests and fixtures built on it: broken-pool.test.ts, health-broken-pool.test.ts, bun-42804-server.ts, broken-pool-detection.ts, and the hook counts pinned by query-logging.test.ts / query-hook-profile.ts

One effect worth stating: production on PostgreSQL/MySQL with query logging off goes back to installing no query hooks at all, the same as SQLite, so bun-query-builder's hook-free fast path applies there again.

Refs #2975