Near-Minimax-Optimal Distributional Reinforcement Learning with a Generative Model