Operator Augmentation for Model-based Policy Evaluation