we still need work on the step retry when gres resources are all busy

a problem still remains with respect to oversubscribing a job's gres allocation at the step level

Get gres configuration changes working.

fix bug in job gres state dup function (failed to append new record on the list)
add the job gres to the step credential

add a function to duplicate a job gres record and fix memory leak in the deletion of a job gres record

to become effective.

and use them to sync node state info on slurmctld restart.

add logic to save/restore a job's gres info and sync it (more work needed to save/restore node state for proper data sync)

disable this test if dedicated nodes is not supported by virtue of Shared=FORCE configuration on the default partition

ok, sbatch now works here.  scontrol, smap and sview are the only non bluegene-linux cross that don't work

It prevents overallocation of gres resources
including a map of specific allocated resources.

Still need to add save/restore of that information,
propagate detains to slurmd, consider topology, etc.