fix the torchvision porting script

[CarloLucibello]

The models left uncommented in the model_list in the port_torchvision.jl script are successfully ported from torchvision
(on julia 1.9 rc2). Is there any need to reupload them on HuggingFace?

For SqueezeNet and DenseNet the script errors due to size mismatches. Maybe wrong ordering again?

close #228

[CarloLucibello]

I guess #206 is a good reason for uploading on HF. Or we want to fix the issue mapping the old weights to the new implementations?

[darsnack]

We can upload new weights, but I don't think running the script is the issue. All these weights already exist on HF, and they can be accessed on master. The issue is that BSON won't load models saved with Julia 1.7+ because TaskLocalRNG is not defined.

So I would say re-running this script with Julia 1.6, then uploading those weights will fix all the issues.