log: prevent deadlock on error when multiple critical errors go to the same buffered sink #72454
Description
This is follow up work to #70330.
This is a likely bug identified in (*bufferSink) accumulator(): when multiple logging calls require a synchronous output concurrently, the errorCh field is overwritten in a data race and can cause a deadlock for the log calls waiting for their error.
Epic CRDB-6669
Jira issue: CRDB-11162