Limit reachability for model-free reinforcement learning of ω-regular objectives.