Push to Hub Callback for non-transformer Keras models

[merveenoyan]

This is a feature request that will improve the UX of Keras audience that will increase the number of non-transformer Keras models on the Hub.

We want a callback that is something similar to below one but for Keras models on Hub. https://github.com/huggingface/transformers/blob/824fd44fc314b0d13b1ef91e8122d35d14af5ad9/src/transformers/keras_callbacks.py#L19

This is relatively easier to implement compared to above one as we don't need to push tokenizer whatsoever (though there might be something like preprocessing layer or something from tf.text that should be pushed, I'll investigate our options, but for now we will only push models).

cc: @osanseviero

[osanseviero]

cc @Rocketknight1 @LysandreJik