Text this: Context-Aware Multimodal Fusion with Sensor-Augmented Cross-Modal Learning: The BLAF Architecture for Robust Chinese Homophone Disambiguation in Dynamic Environments