Text this: GSR-Fusion: A Deep Multimodal Fusion Architecture for Robust Sign Language Recognition Using RGB, Skeleton, and Graph-Based Modalities