Correct some logic so that pack_gorup and pack_grp_bits fields
  are properly set.

$ srun --pack-group=1,2 -l tmp
0: JOBID:43740 HOST:nid00002
0: JOBID:43741 HOST:nid00003
1: JOBID:43741 HOST:nid00003
1: JOBID:43740 HOST:nid00002
2: JOBID:43740 HOST:nid00002
2: JOBID:43741 HOST:nid00003
3: JOBID:43741 HOST:nid00003
srun: error: Task 0 reported exit for a second time.
srun: error: Task 1 reported exit for a second time.
srun: error: Task 2 reported exit for a second time.

srun now parsing for the "-pack-group" option first,
  then processes the environment variables for that pack group,
  then processes the remaining options for the pack group.
srun will still only spawn tasks for one pack group at a time,
  but will at least handled forwarded options from the job
  allocation phase (i.e. sbatch or salloc).

Change (almost) all references to use pointer to structure for
  pack-job support with multiple components

Previous logic could fail on step allocation in larger job allocation

Underlying logic not yet available, just the new option parsing and
  documentation.

Make packjobid, packjoboffset, and packjobidset fields available in squeue
      output.

Test salloc/sbatch job options passed and resulting environment variables.

Improve backfill scheduling algorithm with respect to starting jobs as soon
    as possible while avoiding advanced reservations.
bug 3757

bug 3886

Not necessarily fatal(), but of potential interest when debugging
odd slurmctld crashes. Cannot go where the limit is originally set,
as the logging infrastructure is not avaiable at that point.

Bug 3886.

test1.91 fails with non-default binding

Bug 3581.

Fix for commit 250378c2
  test7.3 was failing without this patch
  bug 3502

test 17.12 was leaving slurm-#.out files around. Explicitly set
  output file to /dev/null and set time limit to 1 minute to avoid
  vestigial jobs.

Handle syncing between siblings (not the origin). Job could have been
canclled or started while the origin was down. The sibling should see
this and remove it's copy if the job is running or cancelled on the
remote cluster.

Handle case where job was cancelled or finished while the origin was
down and the siblings were up.

The controller will keeps job in the job_list until the origin comes
back up and will find out about it then.