(More) Efficient Reinforcement Learning via Posterior Sampling