SpreadWriter: fix flush panic with zero writes

[ploxiln]

fixes #1435

Because the only user of this is nsqd/statsd.go, we could just return without sleeping in this case, and the statsd loop has its own ticker to wait on. But I figured, we might as well make SpreadWriter.flush() always take the desired amount of time, it might be less confusing if this is ever reused, and it's not hard.

(Another interesting idea is to just add a zero-byte buf to the list of bufs if it is empty, but that does result in a zero-byte write syscall, which you can detect with netcat ... it's probably fine, but it's a bit weird.)