Reinforcement Learning with Prototypical Representations