Integration of Microsoft's Kosmos 2.5 OCR and document understanding model

### Requested feature
Can we integrate Microsoft's new Kosmos 2.5 model into Docling? Microsoft released the [Kosmos 2.5](https://huggingface.co/microsoft/kosmos-2.5) document understanding model under MIT license this week. KOSMOS-2.5 can be seen as a successor to models like Donut 🍩 and Pix2Struct by Google. The model can handle 2 tasks: image-to-markdown and performing optical-character-recognition (OCR) on an image. When the model does OCR on an image, it returns both the text as well as the spatial coordinates indicating where the text was found in the image.

Is Komos 2.5 suited for integration with Docling? Would it be of interest to the project?