Learning Decision Policies with Instrumental Variables through Double Machine Learning