Convergence results for an averaged LQR problem with applications to reinforcement learning