bug: pool never recovers after releasing a reserved connection whose backend was terminated.

Hey 👋

Thanks for making JavaScript a better place! 🫶

While working on query cancellation in kysely-postgres-js I stumbled upon this behavior:

1. create a pool of 1.
2. reserve a connection with reserve().
3. run a long query on it, e.g. pg_sleep.
4. terminate the connection's backend in a secondary session via select pg_terminate_backend(pid).
5. the inflight query rejects with CONNECTION_CLOSED as expected.
6. release the reserved connection back to the pool with release().
7. run another query on the same pool. it never settles.
8. we get an uncaught exception:

   TypeError: Cannot read properties of null (reading 'write')
   	at Immediate.nextWrite (.../postgres/src/connection.js:255:22)
   

The pool discarding the dead connection and using another (existing or new) seems like the wanted behavior in this scenario.
Skipping the release(), allows next queries to run.

Possibly related: #1195, #1186.

Happens everytime when using postgres@3.4.9, Node.js v26.1.0, PostgreSQL 18.4.

Reproduce the seven-step scenario with postgres@3.4.9, Node.js v26.1.0, and PostgreSQL 18.4, then inspect postgres/src/connection.js around line 255 and the pool reserve/release path. Done means releasing a reserved connection whose backend was terminated no longer leaves the pool hung or raises the null-write exception, and a subsequent query settles.