Reinforcement Learning with Unbiased Policy Evaluation and Linear Function Approximation