Text this: Enhancing efficient deep learning models with multimodal, multi-teacher insights for medical image segmentation