Machine Learning for Causal Inference: On the Use of Cross-fit Estimators