Catch pg client connection errors + add synchronous crash logging

[CCheng00]

Prevents the recurring idle-in-transaction crash from killing the process. When Postgres terminates a checked-out connection (e.g. a transaction held open past idle_in_transaction_session_timeout while awaiting a slow Twilio/Teams send), the client emits an 'error' event. Previously nothing listened for it on a checked-out client (pool.on('error') only covers idle pool clients), so it became an unhandled 'error' -> uncaughtException -> process exit(1) and a 2-3 min outage while ECS replaced the task.

• db.js: attach client.on('error') in pool.on('connect') (once per physical connection) so the termination is logged ("Database client connection error: ...") and handled instead of crashing the process. The in-flight request still fails and rolls back, but the process stays up.
• crashLogging.js + index.js: install process-level uncaughtException / unhandledRejection handlers that write the stack synchronously to stderr (fs.writeSync), so any future fatal error is captured in CloudWatch even if Sentry's async delivery is lost on exit.

This is a safety net that stops the outages regardless of which code path still holds a send inside a transaction; the remaining in-transaction sends in eventHandlers.js will be moved out separately.