A cleaned up beit3 remap onto vision_transformer.py vit

[rwightman]

Remap of beit3 from #2489 onto vision_transformer.py vit w/ scale norm support added

[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.

[rwightman]

@brianhou0208 so, this is what I ended up with. #2502 started as an experiment, I basically handed things off to Claude to see how well it'd do. What was in that PR was just prompting on top of your initial PR. This is a clean diff of that plus some cleanup.

It started off easy with the architecture changes being picked up decently, but then had quite a bit of back and forth on the weight conversion. It probably would have been similar time to do it myself, but though it was a worthwhile experiment :)

[brianhou0208]

Super cool experiment! Thanks for handling all the back-and-forth and sharing your thoughts.
Also, thanks again for merging BEiT3 into timm :)