When Algorithms Offend: Unpacking the Causes and Cures for User Aversion in AI Recommender Systems