Add docker image and script for submitting eval jobs

[jacob-morrison]

This PR adds support for beaker batch eval jobs. To do this, this includes:

• small tweaks to scripts/run_rm.py to support automatically logging in, plus also saving results in a beaker-readable format
• a docker file defining our docker image (with support for training in a future PR)
• a script to submit eval jobs, and a default yaml file for eval jobs
• a config file containing smart defaults for evaluating different models on a single A100
• short instructions on how to create an updated version of the docker image