Run regression tests on bare metal

[nbren12]

It would be nice to invoke the regression tests on bare metal (e.g. within the nix builds). The tests are also more complex than necessary since they need to manage docker run commands. It also requires that the python requirements be installed outside of the docker container. This means we need to maintain two python environments 1) the host and 2) the docker container. Much of this complexity goes away if we assume that all commands are invoked from within the docker container and pull all the "docker runs" as close to the CI entrypoint as possible.

See ai2cm/fv3gfs-wrapper#272 for example.