Pretrain RL agents using historical flight data.
This process will download the dataset, engineer 20-dimensional features, and train both Linear and Neural architectures.