Byzantine Robust Cooperative Multi-Agent Reinforcement Learning as a Bayesian Game