Upstream training changes to avoid hacky solution BC + PPO for evaluation script

[anthonysimeonov]

We hard-coded some checks for the actor name to use src/evaluate_model.py with residual PPO models. We also just commented out some breaking checks in get_runs that were preventing any of the residual PPO wandb runs from being used.

Let's clean these up and make the necessary upstream changes to the training pipeline/config settings so that everything can be simply inherited from the configs that were created when the training job was launched.

[ankile]

On it, sir!