error is encountered instead of continuing with more job updates.

Fix timout logic in sched/backfill
Modify calls to schedule() to add maximum queue depth, which speeds up the rate at
which jobs can be submitted dramatically. The Full depth is tested by the backfill
scheduler and periodically in slurmctld_background(), only 100 jobs are tested at
the time of each job submit or completion.

    minimize system fragmentation.

Update docs and fix minor bug in step gpu gres env var

More testing is needed, but basic functionality is now available!!!

GresPlugins to GresTypes

speedup under some configurations

	  to properly handle delays

this improves system throughput substantially, but slightly increases
the time window between state saves from 0sec to 5sec

and pull them off the list rather than sorting all pending jobs.

sorting the entire list

    to backfill schedule if slurmctld has more than 5 active RPCs.

make sched_timeout logic the same as that in backfill to avoid message timeouts
in slurmctld_background(), tweak times to spread out the work a bit more