restored the semaphored version to its original simplicity (the -wait_for rule was accidentally checked in earlier)

only run balance_semaphores() automatically when it is absolutely necessary (no running or otherwise runnable jobs)

balance_semaphores() returns the number of unsemaphored jobs, and the second sync() is only done if a nonzero number of semaphored jobs were unsemaphored

simplified the apparent control flow of synchronization subroutines, removed the returned value that is no longer used

Make sure automatic re-balancing of semaphores only happens for pipelines that have specifically asked for it.

internal api cleanup: switched to more intuitive set_and_update_status/check_in_worker/check_in_job calls

added an interface method complete_early() that should be used instead of calling incomplete(0) and die

make sure beekeeper quits when looping in -logic_name XXX mode and there are no more jobs in this analysis -- thanks, Irina!

attempt to re-introduce the Workers that have never specialized and therefore do not have a real analysis_id,
yet have claimed resources and so should influence the stats

Conflicts:
	scripts/generate_timeline.pl

a more correct treatment of resource_class_id (first call is from a specific worker, and then from the analysis)

schema patch_of_a_patch: ignore possible collisions of meta_keys in meta that are no longer tolerated in pipeline_wide_parameters.