Low-Rank Representations Meets Deep Unfolding: A Generalized and Interpretable Network for Hyperspectral Anomaly Detection