Review for NeurIPS paper: Upper Confidence Primal-Dual Reinforcement Learning for CMDP with Adversarial Loss