C-GAIL: Stabilizing Generative Adversarial Imitation Learning with Control Theory