Multi-agent RL for Petting Zoo

[lhnguyen102]

closes #103

MR includes:

• Reinforcement learning with human in the loop learning for petting zoo
• Reinforcement learning with human feedback for petting zoo
• Self-training RL for petting zoo

Note that we use PPO for these three above-mentionned types of learning

[lhnguyen102]

@cloderic I've tried to merge AtariEnvironment with ClassicEnvironment, but due to the differences in action types and the way we handle next player, I've decided not to do the merger at this PR. Instead, I would like to focus on implementing TAMER and ensuring that self-playing works, which is urgent, given our due date to AAMAS

[cloderic]

Let's merge it, there's still some stuff to improve here and there but it's gonna be easier once merged.