Reinforcement Learning with General Value Function Approximation: Provably Efficient Approach via Bounded Eluder Dimension