Integrate with huggingface

[not-lain]

🔥 Summary

This pr introduces better integration with huggingface (and for local usage as well)
allowing users to use from_pretrained push_to_hub, and save_pretrained to directly load and save their models with ease (similar to the transformers library )

📖 📷 Additional context

• Any new dependencies added? N/A
• Any breaking changes introduced? backward compatibility is ensured
• Screenshots?
  

🛠 Testing instructions

I have prepared this notebook to help you migrate your current checkpoint to the supported format: https://colab.research.google.com/drive/1NdS2sTgJCSP2opLU3rC5BaG_npPDRv0S?usp=sharing

Important

It is essential that you run the notebook above to equip your repo with the new changes

once your repo has been equiped users can directly load your checkpoint with just 2 lines of code

from cube3d.inference.engine import Engine # or EngineFast
model = Engine.from_pretrained("not-lain/cube") 

✅ Checklist

• Provide testing instructions
• Update relevant documentation

Do not hesitate if you have any reviews on the pr or any follow ups.

Best regards,
Hafedh Hichri
Huggingface Fellow