LPLgrad: Optimizing Active Learning Through Gradient Norm Sample Selection and Auxiliary Model Training