Robot Policy Learning with Temporal Optimal Transport Reward