Text this: A multimodal visual–language foundation model for computational ophthalmology