Support 'Devstral 2 (+Small)' models from mistral.ai

[chris-hatton]

Mistral.ai have done it again; releasing 'Devstral 2' in regular (123B) and Small (24B) variants.

Being a popular use-case from a popular vendor, llama.cpp support would surely be welcome.

Here are the HuggingFace model cards:

• Devstral 2
• Devstral 2 Small

[thor171]

If it's not supported, why do I use it since the release of the Unsloth quants?
I don't have problems using it, except for the low performance...