Documentation of GPU setup

We want to enhance the out-of-the-box experience for GPU users and advise for best setup. In this issue we collect a few bullets and ideas for the new docs page.

(in no binding order)
- docs for launching `vllm serve granite-docling` and how to use it in Docling. Best parameters for concurrency, etc
- best/recommended setup on Mac (usually mlx), Linux, Windows
- print warning in the CLI if a non-optimal setup is used