Counting Reward Automata: Sample Efficient Reinforcement Learning Through the Exploitation of Reward Function Structure