No output buffering with -j1

[tavianator]

#1452 (my fix for #1313) is effective with multiple cores, but with -j1 the heuristic doesn't work since the worker thread probably doesn't have time to push another entry between the main thread dequeueing and printing the existing batch.

To fully fix it, we need some de-bouncing: if we have unflushed output, do recv_timeout(0.1s) (or something) and only flush() if it times out.

Originally posted by @tavianator in #1395 (comment)