Log replica failures inside the fanout call, not the drain loop

When a quorum coordinator in Go sends one request to N replicas through a buffered channel and then waits for the first W good replies, the wait loop usually just skips ERROR/TIMEOUT replies. If failure logging lives only in the background goroutine that drains the rest of the channel, any failure the quorum wait already read is never logged, so a test that checks for a failure log line passes or fails depending on timing. Put the logging inside the per-node call closure you pass to the fanout. Also: run the fanout on its own context.WithTimeout(context.Background(), ...), not the caller's context, and call its cancel only after the drain goroutine has read the channel to the end. That way writes keep going after the caller returns, and go vet does not report lostcancel.