Add a job flag to indicate when the EpilogSlurmctld us running
and don't purge the job record until it completes. This lets
the EpilogSlurmctld requeue the job and otherwise manage it.
bugs 635 and 636

Change warnings about slurm.conf configuration problems only for
1) user root, who is in a position to do something about this OR
2) other users only if they use the "-vv" (very verbose option)
bug 639

bug 640

on any of the commands.

application environment the PMI2 library will connect the the PMI2
server (slurmstepd) instead of using the provided PMI_FD socket.

Conflicts:
	RELEASE_NOTES

Without this, an xfree could be called for an arbitrary memory location

Rather than continuously retrying a step create for suspended jobs,
add a sleep with exponential backoff

If a job is suspended, log the step create failure using debug
rather than info in slurmctld

Conflicts:
	src/plugins/switch/nrt/nrt.c

Cache results for major performance improvement.
bug 636

The test for NRT_NULL_MAGIC failed to capture some problems
if the pointer to the structure was NULL. This is an ammendment
to commit 2a55aa0b

If a job request explicitly requests a GRES count of zero and that
is not the last GRES in the slurm internal data structures, the job
request will be rejected.
bug 633

Conflicts:
	src/plugins/switch/nrt/nrt.c

Perhaps should also look into doing this for nodeinfo and libstate

this would cause pmd's to hang.