Refactor train.py and cfg to allow hydra tab completion

[tylerlum]

Goal

Add Hydra tab completion for train.py to help users write commands more quickly, rather than needing to constantly check the config yaml files to remember the parameter names (eg. I kept forgetting if it was num_envs or numEnvs)

^ Example gif showing what the tab completion looks like

How To Use

To set up, run:

eval "$(python train.py -sc install=bash)"

Then simply type python train.py and then the tab completion will work. You can start typing a parameter name (eg. python train.py num_) and press tab to either autocomplete or show possible completions.

Low-Level Details About Changes

• Needed to move many train.py imports into the function because these imports caused prints to terminal that made the tab completion setup not work correctly
• Needed to explicitly add version_base 1.1 to avoid warning messages that made the tab completion not work as nicely (https://hydra.cc/docs/upgrades/version_base/). This matches the behavior from before (not setting the version base makes it default to 1.1, so this is just more explicit)
• Needed to add the explicit _self_ at the end of the default list in hydra to avoid warning messages that made the tab completion not work as nicely (https://hydra.cc/docs/upgrades/1.0_to_1.1/default_composition_order/). This matches the behavior from before.
• Needed to add override in config file to avoid warning (https://hydra.cc/docs/1.2/upgrades/1.0_to_1.1/defaults_list_override/)

Relevant Screenshots from Hydra Documentation

^ currently using version_base 1.1, so set this explicitly

^ using version 1.1, so put _self_ last

^ add explicit override term