Download torchvision models

[wdhorton]

Right now in order to use pre-trained torchvision models (as well as libraries using torchvision, like fastai), you need to import the weights as another dataset (like https://www.kaggle.com/pytorch/resnet34) and then write some code to copy it into the right place in the filepath. The code looks like this (from this kernel):

cache_dir = os.path.expanduser(os.path.join('~', '.torch'))
if not os.path.exists(cache_dir):
    os.makedirs(cache_dir)
models_dir = os.path.join(cache_dir, 'models')
if not os.path.exists(models_dir):
    os.makedirs(models_dir)
!cp ../input/resnet34/resnet34.pth /tmp/.torch/models/resnet34-333f7ec4.pth

I think that we can help people avoid having to work around it in this way--since the Dockerfile already has downloads for other tools (like spacy and nltk), I think it would be helpful to also download some pretrained image models so people can get started in a more straightforward manner. And since the image is already pretty large (~16.8GB), it shouldn't make a significant impact on the size.