- more carefully counting the running and exited workers

	- "< max threads running" is a warning, not alert

	- restored hash entry check against fd, but not if stale (-1)
	- removed membar after client bit not set (didn't help)
	- intial data send fail (before reconnect) is a warning (not alert)

	set -- false fd's tend to come after this, maybe data are
	stale
	- dump the data table before abort