Parallel Reinforcement Learning with Linear Function Approximation