Leave the matter of document parsing to PaddleOCR-vl
RP-StructureV3 for structured document pipelines.