MDP Homomorphic Networks: Group Symmetries in Reinforcement Learning