Reviews: Verifiable Reinforcement Learning via Policy Extraction