Add PP-HGNet and PP-HGNetV2

[seefun]

PP-HGNet is a high-performance backbone network developed by the PaddleCV team that is more suitable for GPU devices. Compared with other SOTA models on GPU devices, this model has higher accuracy under the same latency. (From PaddlePaddle Clas repo, but I think it's a variant of VovNet)

https://github.com/PaddlePaddle/PaddleClas/blob/release/2.5.1/docs/en/models/PP-HGNet_en.md
https://github.com/PaddlePaddle/PaddleClas/blob/release/2.5.1/docs/zh_CN/models/ImageNet1k/PP-HGNetV2.md

The implement and pretrained weights are adapted from paddle.

Pretrained weights converted from origin PaddleClas repo: Google Drive

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.