Adds Ability for Policies to Play Policies Or Humans

[decodyng]

• Refactors dropdown to explicitly select players for Player0 and Player1, which can be either human or agent. As part of this, hat colors no longer map to player/non-player, but just correspond in a static way to the index of the player, which is indicated in the text above the dropdown.
• If both are agents, have agents play with each other
• If one is a human, act normally
• If both are humans, yell because we don't have the keyboard listeners for that

[decodyng]

Re including test_traj.json: I think the best thing to do would be to include a set of test_trajectories that represent non-arbitrary things, but that might need to wait until you guys wrap up NeurIPS responses for Micah to have the time. In the absence of that, I lean towards keeping it in just so it works at all out of the box.

Re incompatible models and layouts: I suspect that there would be some error message on the backend/console, but agree that it wouldn't show up in the front end as currently coded. I'll add it as an issue!

[rohinmshah]

LGTM, feel free to merge. (The conflict looks easy to resolve.)