Finite-Time Regret of Thompson Sampling Algorithms for Exponential Family Multi-Armed Bandits