Support GPU requests

[treydock]

It would be useful to have ability to directly request GPUs via the various scheduler adapters.

SLURM:

--gpus-per-node=N

Torque:

-l nodes=X:ppn=Y:gpus=N

Kubernetes: https://kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/

resources:
  limits:
    nvidia.com/gpu: 1
  requests:
    nvidia.com/gpu: 1

[ericfranz]

And then presumably OodCore::Job::Script would accept as an argument in its initializer: gpus_per_node: <Integer or nil> so the default value is nil (not set).