Currently Pegasus doesn't support GPU exposure to containerized jobs.
We should include basic support for nvidia gpus, as the first step.

In the case of singularity we should append the cli argument "--nv", as described here: https://sylabs.io/guides/3.5/user-guide/gpu.html

For docker we should append the cli argument "--gpus all", as described here:
https://github.com/NVIDIA/nvidia-docker