feature(picture-description): inject document context (caption, surrounding text, classification) into picture description prompt

### Requested feature
Image description backends send every image with the same static prompt, blind to its
context within the document.

We can leverage the surrounding text, caption, or classification label to ground the
description. I will work on this and send a PR, it should be a quick win since all
the information is already available in the `DoclingDocument` at enrichment time.