On Reward-Free Reinforcement Learning with Linear Function Approximation