Would you consider adding a batch prediction for the layout_predictor.py if I made a PR for it?

I'm using the models separately outside of the `docling` library. Batch inference is signifcantly on GPUs and in particular if I wish to infer the laout of over 100 pages at once.

If I get the go ahead. I can make a nice little for it, please lmk!

Thanks for the brilliant work on this project! 🤗 