Fitting Double Machine Learning Models with Instrumental Variables and Potentially Heterogeneous Treatment Effect