Text this: Predicting fatigue limits of defective A356-T6 and A357-T6 cast aluminum alloys using a hybrid empirical–machine learning approach