How Reliable Are Out-of-Distribution Generalization Methods for Medical Image Segmentation?