nsq_to_file: with a large max-in-flight some messages unprocessed during a SIGTERM

[danbf]

Root cause: Go's select chooses randomly when multiple cases are ready simultaneously. At shutdown with 1000 max-in-flight:

The last handler goroutine sends its message to logChan (buffer=1, so it succeeds immediately and the goroutine returns)
go-nsq sees all handlers have returned → closes StopChan
Now both StopChan and logChan are ready at the same time
If select picks StopChan, the message in logChan is abandoned — never written to disk, never Finish()'d
The higher the max-in-flight, the more often this happens because there's more concurrency in the final draining.

Fix: The StopChan case now runs a non-blocking drain loop over logChan before setting sync = true. Since StopChan only closes after all handler goroutines return (and each goroutine sends exactly one message and returns), the buffer holds at most one message at that moment — but the loop is written generically with a labeled break for safety.

NoteL This may not be important as re-queuing should prevent messages from getting dropped.