Display nodes to be used in the scheduling of a pending job.

Add support for allocation of GRES by model type for heterogenous systems
(e.g. request a Kepler GPU, a Tesla GPU, or a GPU of any type).

not been processed by the scheduler yet.

No change in logic

Modify srun to report an exit code of zero rather than nine if some tasks
exit with a return code of zero and others are killed with SIGKILL. Only an
exit code of zero did this.

The fprintf(stderr will generally fail as stderr will usually be closed.

Cache job information and drop out of loop after finding the
desired array task. This improves performance a fair bit.
bug 684

Rather than attempting a scheduling loop as frequently as every
1 second, wait 3 seconds for performance reasons.

set it in the job env otherwise the exec() fails.

Rather than immediately invoking an execution of the scheduling logic on
every event type that can enable the execution of a new job, queue its
execution. This permits faster execution of some operations, such as
modifying large counts of jobs, by executing the scheduling logic less
frequently, but still in a timely fashion.

If you have multiple partitions the output from

sinfo -o "%D %F"

would have unexpected results, hardly ever correct.

was originally testing for NULL value in one field and printing
another, which could possibly result in an invalid memory reference.

This just adds logic so slurmd can read "type" from a gres.conf
and ship it to slurmctld.

Conflicts:
	META
	NEWS