Combining Reinforcement Learning and Optimal Transport for the Traveling Salesman Problem