Minimizing passenger waiting time in the multi-route bus fleet allocation problem through distributionally robust optimization and reinforcement learning