High-Dimensional Optimal State-Feedback Mapping Using Deep Neural Networks For Agile Quadrotor Flight