t-DGR: A Trajectory-Based Deep Generative Replay Method for Continual Learning in Decision Making