Energy Efficient Edge Computing: When Lyapunov Meets Distributed Reinforcement Learning