Dynamic wagon assignment and reassignment for pick-up and delivery requests: A deep reinforcement learning framework to address uncertainty in locomotive availability