[HUB] add SOTA image compressor for classification.

[YannDubs]

I'm adding the CLIP self-supervised compressor from FAIR + Vector Institute.

This allows compressing any dataset orders of magnitude better than with JPEG (e.g. 1000x for ImageNet) without decreasing downstream classification accuracy :)

[netlify[bot]]

:heavy_check_mark: Deploy Preview for pytorch-hub-preview ready!

:hammer: Explore the source changes: 43168ba0a1b316fabebbb06184d7494200fa2e5f

:mag: Inspect the deploy log: https://app.netlify.com/sites/pytorch-hub-preview/deploys/60e5a088aa69960007e35c77

:sunglasses: Browse the preview: https://deploy-preview-213--pytorch-hub-preview.netlify.app

[YannDubs]

The circleci failure comes from another script (Running pytorch example in pytorch_fairseq_roberta.md), not this one.