Extract Receipt

Extract purchased line items from a receipt photo/PDF.

Synchronous — one image/page, fast enough for a single request (unlike BOM import's multi-drawing background job). Stores the file to the project bucket too, so the frontend can tag it as item_documents on whichever extracted lines the user chooses to add.

Returns extracted data only — no DB writes. The frontend shows a review screen (vendor/date/payment + editable item rows) before anything is added, since vision extraction from a photo can misread a price or name.