A Distributional Analysis of Sampling-Based Reinforcement Learning Algorithms