[Integration] Adding a better OCR vision model

[yoeven]

Hey there! We love the approach of using a vision model to generate markdown but it isn't full proof all the time. So we trained a vision model (vLLM) with traditional OCR and now we get more consistency and cleaner data output and native support for PDFs, image and supports response structure like JSON & Markdown. It's called JigsawStack vOCR.

If you think it makes sense, happy to create a PR that adds this integrations as an option between a default LLM or the vOCR model. Let me know what you think :)