Uniform-PAC Bounds for Reinforcement Learning with Linear Function Approximation