Language Meets Vision: A Vision–Language Guided Multimodal Framework for Explainable Industrial Anomaly Segmentation