Learning to Solve Multiple-TSP with Time Window and Rejections via Deep Reinforcement Learning