Switch transcription to the published cv11-v5 PT-BR conversion

Make Q&A usable during processing; stream OCR per image

Read PDF attachments first; skip webp images