Create scripts to run things on NERSC/Frontier/Aurora

[anagainaru]

With explanation on how to change processes/nodes/memory requirements

[rz4]

#78 and #67.

These are very barebones. I have a readme per deployment environment that describes how to get to the point where you can submit a job that runs the MNIST example successfully.

The next step is to make user we can submit jobs that utilize more than 1 GPU.

The number of process/nodes/memory will depend on the example. We can use Prometheus as the example on how to configure jobs based on the requirements of the model, and update the deployment READMEs as needed.

[rz4]

I have yet to get my account on Aurora.

Permutter and Frontier deployment are streamlined and ready for single GPU runs.

Need to test DDP training next.