[FEATURE] Add Offline batch generation for open models with EXXA API

[etiennebalit]

Is your feature request related to a problem? Please describe.
Batch generation has cost and efficiency advantage over streaming APIs and works well for synthetic data generation.

Distilabel now integrate OpenAI Offline Batch Inference but open models like Llama 3.1 still require hosting the model yourself or going through a streaming API provider.

Describe the solution you'd like
Add a new ExxaLLM whichs implements offline_batch_generate as OpenAILLM for open models on EXXA (starting with Llama 3.1).

Additional context
We're happy to submit a PR soon.

[gabrielmbmb]

Hi @etiennebalit, it would be awesome if you could submit a PR!