pg-pool: enable TCP keepalive and retry once on connection-drop errors

[16francej]

Problem

Managed Postgres poolers/proxies can drop idle or in-flight connections (rebalancing, failover, idle-connection reaping); flaky networks do the same. When a connection dies mid-query, pg surfaces:

• Connection terminated unexpectedly / Connection terminated
• ECONNRESET / EPIPE socket errors
• 57P01 (terminating connection due to administrator command)

and the request fails with a 500 even though the statement never ran anywhere.

Repro: kill the Postgres backend mid-query → the caller gets a 500 with Connection terminated unexpectedly.

Fix

• Set keepAlive: true on the pg.Pool so dead peers are detected promptly and idle connections are less likely to be reaped by middleboxes.
• In the pool's simple-query path (query()), retry once on a fresh client when the failure is a connection-drop error, via a new exported isConnectionDropError(err) helper. Any other error still throws immediately.

The retry is deliberately single-shot and limited to the pool-level query() path, where each attempt checks out a fresh client from the pool (the dropped client is discarded by pg), so the retry never reuses a dead connection. Transactions (withPgTransaction) are untouched — replaying those is not reuse-safe.

Tests

• New unit test covering isConnectionDropError (positive shapes + non-drop errors + non-object inputs).
• Existing pg-pool, pg-ca-options, and persistence-init-retry suites pass; tsc --noEmit clean.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}