Fix state machine to support corner cases when fan-in message for the next Fence appears before fan-out message completing the current Fence. This was observed on Orion as well as on containers environment in cases where Fences was light but the number of process was reasonably high.

Change the ordering of operations during fan-in. We will need this for the improved state machine. This changes are verified to work.

Optimization of fan-in collective stage. Sense parent by sending him a zero-sized message before sending him a junk of data. This helps to resolve cases where parent slurmstepd in collective haven't been yet bootstrapped.

Fix EIO: try to process message right in the place if possible instead of always switching to the AIO event handling.

This reverts commit 76098daf.

bug 2171

Decrease parallelism in job cancel request to prevent denial of service
    when cancelling huge numbers of jobs.
bug 2256

Prevent triggering gang scheduling within a partition if configured with
    PreemptType=partition_prio and PreemptMode=suspend,gang.
The essence of this fix is to change a "<=" to "<" in cons_res/job_test.c:
-               if ((p_ptr->part_ptr->priority <= jp_ptr->part_ptr->priority) &&
+               if ((p_ptr->part_ptr->priority < jp_ptr->part_ptr->priority) &&
but logic was also added to insure that a partition configuration with
PreemptMode did not override PreemptType != partition_prio.
bug 2232

Avoiding conditional directives that break statements

Conflicts:
	src/slurmctld/job_mgr.c

No changes to logic

Previously an error() would be logged when the attempt to open the job
script using the new directory format failed but the successive fallback to the
old directory structure was successful, leading to confusion when troubleshooting.

Move emitted warnings to debug(), and only error() after failing to open in both
directory structures. Add a note about backwards compatibility to both functions -
we cannot remove these fallbacks as directory structure for pending jobs does not
change on Slurm version update, and people may need to chain multiple version
update together to get to a current slurm version which would correctly update
slurmctld state files but leave pending jobs in the old directory structure.

Bug #2244.

If a job is requeued while in the process of being launch, remove it's
    job ID from slurmd's record of active jobs in order to avoid generating a
    duplicate job ID error when launched for the second time (which would
    drain the node).
bug 2240

In slurmctld log file, log duplicate job ID found by slurmd. Previously was
    being logged as prolog/epilog failure.
bug 2240