Document question answering (image + optional word_boxes)
Body
Request for document (image) question answering pipelines.
Image of the document (PIL-like object, base64 string or binary upload). Required.
Question to ask about the document. Required.
Optional list of (word, box) tuples or normalized bounding boxes to provide OCR words/boxes.
Number of top answers to return.
Stride size for splitting long documents into chunks with overlap.
Maximum length (in tokens) for returned answers.
Maximum total sequence length (context + question) in tokens.
Maximum question length after tokenization.
If true, allow returning an 'impossible' / no-answer result.
Language hint for OCR; default is English if omitted.
Extra configuration flags to pass to Tesseract OCR (if used).
Maximum time in seconds to wait for external image fetching/processing.

