[FEATURE] Support variable input size of maxvit and coatnet

Will it be possible in the future to support variable input sizes for maxvit and coatnet?

I am experimenting with adapting various models of timm to self-supervised learning such as DINO and iBOT.
When learning these, it is necessary to accept inputs of (224,224,3) and (96,96,3) at the same time.

[The model](https://github.com/bytedance/ibot/tree/main/models) here seems to support variable sizes. If you're not going to break the implementation, I'd love to see you add support for variable sizes.

Start by locating the MaxViT and CoAtNet model implementations and reviewing how their input sizes are handled. Check both (224, 224, 3) and (96, 96, 3) inputs, including simultaneous use for self-supervised learning, and verify that existing model behavior is not broken.

Written by the indexing model from the issue text.