Update torchvision.tranforms version to torchvision.transforms.v2

[lixit]

According to https://pytorch.org/vision/stable/transforms.html#v1-or-v2-which-one-should-i-use

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[tfriedel]

There's more v1 transforms in

pytorch-image-models/timm/data/transforms.py

Line 399 in 45eb69d

class ResizeKeepRatio:

Maybe in other places as well, haven't checked.

Also v2 support mixup/cutmix. Maybe it's not so trivial to modify timm to use them, but I guess they will be faster.