feat(petals): add backend

[mudler]

Description

This PR fixes #1262

It adds petals (https://github.com/bigscience-workshop/petals) support - it does not support token generation yet, but is a first attempt on supporting p2p based backends. really interesting as this unlock usage of LocalAI in very low or embedded environments.

Maybe time to get arm64 workers so we can build images faster? cross-compiling on GH was impossible to handle anymore

Example model config:

name: petals
parameters:
  model: petals-team/StableBeluga2
backend: petals

curl call:

curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
     "model": "petals",
     "messages": [{"role": "user", "content": "Once upon a time..", "temperature": 0.1}]
}'

[netlify]

✅ Deploy Preview for localai canceled.

Name	Link
🔨 Latest commit	1c09529
🔍 Latest deploy log	https://app.netlify.com/sites/localai/deploys/6564e73b2eba8c00080ff5af