Enable docker model runner

[ribalba]

It is now possible to run AI models directly through the docker model runner. GPU support etc comes natively. We should extend the GMT to allow energy benchmarking of these runs.

https://docs.docker.com/ai/model-runner/api-reference/

We should also check if it is possible to somehow get single prompts measured. This should also become a blog post.

[ArneTR]

I wonder what the benefit is atm. It seems like it is even a bit more verbose than our implementation with running the OLLAMA container.

If that really becomes standard I think it would be nice to support it. Otherwise I believe it is better to not open another security vector and implement this feature.
Our implementation atm allows to restrict certain action to only "premium users" and others to public ones, like for instance downloading a new model etc. - Chances are this is more complicated to adapt when we use the published API endpoint inside the container.

Another problem that I see is that our infrastructure currently does not support it.

See: https://docs.docker.com/ai/model-runner/#requirements

NVIDIA drivers 575.57.08+

Our machines currently only support NVIDIA kernel driver version 535.261.03
See: https://docs.docker.com/ai/model-runner/#requirements

Happy to leave this issue open for processing in the future