Exponential Hardness of Reinforcement Learning with Linear Function Approximation