Mo\"ET: Interpretable and Verifiable Reinforcement Learning via Mixture of Expert Trees