add MaxViT [TF]

### Model description MaxViT: Multi-Axis Vision Transformer is one of the nice papers of late 2022 which is also published in [ECCV 2022](https://arxiv.org/pdf/2204.01697v4.pdf) by Google AI. * This paper introduces a new attention module called "multi-axis attention" which consists of blocked local and sparse global attention for efficient and scalable spatial interactions on arbitrary input resolutions. * It demonstrates superior performance on various vision tasks including image classification, object detection, and so on. I think it would be nice to have it on Hugging Face. I would be happy to contribute it on Hugging face. cc: @alara @NielsRogge ### Open source status - [X] The model implementation is available - [X] The model weights are available ### Provide useful links for the implementation Code and Weights: * TF1: https://github.com/google-research/maxvit * TF2: https://github.com/leondgarse/keras_cv_attention_models