Reducto originally combined traditional computer vision to preserve document structure with vision language models acting as an error-correction loop