Adversarial Inverse Reinforcement Learning for Mean Field Games