Transfer Learning for Classification under Decision Rule Drift with Application to Optimal Individualized Treatment Rule Estimation