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.

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.

This allows the origin to be able to sync up jobs after it has been
down.

(void *)(intptr_t)0 is treated as a NULL

_cleanup_removed_origin_jobs() could have been called without ever being
part of a federation.

Job could have been requeued if the nodes failed.

bef69448 was fixed/changed so that slurm_addto_char_list() would now
add an empty string to the list if no constraints or clusters were
given. The code was expecting an empty List previously.

Like sview it wasn't mapping the job's node indexes to the correct
nodes since federated nodes are merged into one array.

while the cluster is down. The cluster will figure out what changed
after starting up or after resuming from using the cache.

When the controller starts up and the dbd is not up it waits until the
dbd comes up. At this point, the controller needs to find out if
anything has changed in the federation (e.g. other clusters or self
removed from cluster).

When a non-origin cluster is removed:
- running jobs remain - fed_details removed so it can't call home.
- origin cluster removes tracking job for running jobs
- pending jobs are removed.
- pending srun/sallocs don't get notified.
- other clusters remove removed cluster from viable and active sibs

When an origin cluster is removed:
- all pending jobs are removed from all clusters that had job.
- pending srun/sallocs are notified of termination
- running jobs remain.

The SLURM_ID_HASH used for Cray systems has changed to fully use the
entire 64 bits of the hash.  Previously the stepid was multiplied by
10,000,000,000 to make it easy to read both the jobid as well as the
stepid in the hash separated by at least a couple of zeros, but this
lead to overflow on the hash with steps like the batch and extern step
where they used all 32 bits to represent the step.  While the new method
ruins the easy readability it fixes the more important overflow issue.
This most likely will go unnoticed by most, just a note of the change.

Bug 3815

It would be nice to figure out a way to remove the check for version all
together but I (Danny) couldn't figure out how that would be done since we
need to know which libs/headers to use and on systems with multiple
installed and no 'lua' lib (Ubuntu) you have to use the PKG_CHECK_EXISTS
to set up the pkg name for PKG_CHECK_MODULES or you don't get things set up
correct when trying to link.