Determining optimal strategies for personalized atrial fibrillation treatment in intensive care unit patients using a deep learning-based causal inference approach: rhythm and/or rate control