Non-Adversarial Inverse Reinforcement Learning via Successor Feature Matching