Curriculum Reinforcement Learning via Constrained Optimal Transport