Dynamic Airspace Reconfiguration With Deep Reinforcement Learning