Text this: A multi-modal multi-branch framework for retinal vessel segmentation using ultra-widefield fundus photographs