Efficient and Adaptive Posterior Sampling Algorithms for Bandits