WIP - Add GPU resource requests/limits

[sl1pm4t]

FYI @abagshaw

I haven't been able to test successfully yet. On GKE, the pod stays in pending state forever because the scheduler thinks there aren't any GPU's available. This is despite creating the cluster with GPU support etc. However, assuming your cluster is setup correctly, I think this should work.

Due to Terraform resource attribute naming restrictions (no periods or slashes) the resource name is nvidia_gpu instead of the Kubernetes documented name of nvidia.com/gpu.

Example:

      resources {
        limits {
          nvidia_gpu = 1
        }
      }

Fixes #37