State Chrono Representation for Enhancing Generalization in Reinforcement Learning