SVRG for Policy Evaluation with Fewer Gradient Evaluations