Stable Reinforcement Learning with Unbounded State Space