Improve Hugging Face integration

[NielsRogge]

This PR showcases how a VGGSfM model can be safely pushed to the hub using the PyTorchModelHubMixin. It leverages safetensors to safely store the weights.

By inherting from this class, the following automatically works:

from vggsfm.models.vggsfm import VGGSfM

model = VGGSfM.from_pretrained("nielsr/vggsfm")

# push a trained model to the hub
model.push_to_hub("nielsr/my-new-vggsfm-model")

# save locally as safetensors
model.save_pretrained("path_to_local_folder")

The model is here for now: https://huggingface.co/nielsr/vggsfm, but we can push it to the facebook org instead. As can be seen, download stats also work, along with automated tags ("image-to-3D") and model card:

The entire init arguments are serialized as well into a single config.json: https://huggingface.co/nielsr/vggsfm/blob/main/config.json.

I pushed the model using this Colab notebook.