Draft: mmlu translated professionally by OpenAI

[giuliolovisotto]

Following up from #2305

@baberabb

• I grouped by language in subfolders
• I created a group per language
• I mostly copied the existing mmlu/default so I dropped some options to do cot/generative/fewshots.

I'll try to run something like llama 70b in the next days to make sure numbers make sense.

[giuliolovisotto]

Openai uploaded the dataset on hf, I'll use that.