Computational Flight Control: A Domain-Knowledge-Aided Deep Reinforcement Learning Approach