[T-071] figmark can't take a standalone image as input — a raster image is rejected, though its whole engine is figure interpretation + OCR

[Ztein]

Canonical ticket: docs/tickets/T-071-standalone-image-input.md

figmark accepts documents (PDF, EPUB/native, Office) but not a standalone raster image. A PNG/JPG/… is not in SUPPORTED_FORMATS (input_formats.py), so the CLI and /v1/convert reject it, and over /v1/ocr a document.image_url is resolved to bytes but then fails the format gate → 415 — even though image_url is a documented Mistral OCR input type and a raster image maps exactly onto machinery figmark already has. This is the T-052 deferred item, now with its own ticket.