Enhancing Clinical Text Classification via Fine-Tuned DRAGON Longformer Models