Maximum-Likelihood Inverse Reinforcement Learning with Finite-Time Guarantees