Minimax-Optimal Multi-Agent RL in Zero-Sum Markov Games With a Generative Model