OVT-Net: Semantic segmentation of gastrointestinal cancer using an optimized vision transformer model with explainable AI