FP-IRL: Fokker-Planck-based Inverse Reinforcement Learning -- A Physics-Constrained Approach to Markov Decision Processes