Solving the Traveling Salesperson Problem with Precedence Constraints by Deep Reinforcement Learning