Regularization-Adapted Anderson Acceleration for multi-agent reinforcement learning