Pix2Struct: Screenshot Parsing as Pretraining for Visual Language Understanding