This is follow up work to #70330.

Currently the bufferSink flush goroutine will wait until it has written all its incoming messages before shutting down.
If the underlying output sink is exccessively slow (or blocks outright), this makes the goroutine block for too long.

We want a timeout be added in there during shutdown, so that if the messages cannot be flushed under a reasonable time the goroutine abandons the flushes and simply reports that it has given up.

Epic CRDB-6669

Jira issue: CRDB-11167