Text this: Cross-domain zero-shot semantic segmentation for unstructured environments via EVA-CLIP model, ensemble prompt engineering, and optimized text-image matching.