On the Tension Between Optimality and Adversarial Robustness in Policy Optimization