Add multi-node scaling for match template program

[mgiammar]

Overview

Adds new feature for running the match template program across multi-node, multi-GPU systems by using the torch.distributed package. New file run_distributed_match_template.py is necessary for setting up the distributed environment and simplifying codepaths for the more common single-node, multi-GPU runs.

Todo before merging

• Check that program executes properly on a distributed environment
• Remove debugging print statements
• Add click options to the Python script
• Include basic docs on distributed match template

[mgiammar]

Closes #89

[mgiammar]

@jdickerson95 I'm using development_v1.1 as a quasi-staging branch for the new backend changes, but I think this PR is adding enough that it warrants a brief review to ensure what I'm proposing to add makes sense.

[jdickerson95]

Just remove the unnecessary .cpu() (unless they are necessary) and then I'm happy for it to be merged.