Do Vision Foundation Models Enhance Domain Generalization in Medical Image Segmentation?