Text this: Visual information extraction from documents via classification-guided large vision-language models.