being scheduled to run again.  This would erase the previous job info
that was still needed to clean up the nodes from the previous job run.

Move calls to core_spec_g_set/clear from slurmd to slurmstepd.
Pass count of specialized cores from slurmd to slurmstepd.

change the core_spec_g_set and clear functions to take container
ID as an argument

Move core_spec_g_suspend() and core_spec_g_resume() from slurmd
to slurmstepd daemon per Cray request.

The function arguments are unchanged, this splits the set into
four functions.

a step using --cpu-freq completes.

Conflicts:
	NEWS
	doc/html/team.shtml

Set GPU_DEVICE_ORDINAL environment variable.

Splitting the job's shared field into whole_node and share_res
in order to support gang scheduling of jobs with exclusive node
use broke the logic used to support advanced reservations containing
specific cores.

Conflicts:
	META
	NEWS
	contribs/perlapi/libslurm/perl/Slurm.xs

Previous logic would try to pick a specific node count and on a
heterogeneous system, this would cause a problem. This change
largely reverts commit a270417b

beside the numerical values.