Mixed Autonomy Traffic: A Reinforcement Learning Perspective