CLIP doesn't get installed properly

[PhilippKaniuth]

pypi doesn't like the direct github dependency for CLIP, see here for a corresponding pypi issue. As a result, thingsvision is installed w/out CLIP support which will throw an error for a user wanting to extract from a CLIP model.

@LukasMut stated that PEP might be the actual problem and no matter how the dependency is adapted pypi refuses to take it.

@Alxmrphi suggested that it might be easier to initiate the download when a user actually needs CLIP for the first time (similar to how one checks for local stored weights in pretrained models in torchvision).

The workaround for this is asking every user to do pip install git+https://github.com/openai/CLIP.git after having executed pip install --upgrade. See commit dc5b376f44fb73c2a1c93c91bb911d012ad64425 that added this workaround to the README which would need to be undone when this issue is resolved.