Prior-dependent analysis of posterior sampling reinforcement learning with function approximation