The Unintended Consequences of Discount Regularization: Improving Regularization in Certainty Equivalence Reinforcement Learning