[DRAFT] Keras v3 [DRAFT]

[felixdittrich92]

• PR to keep track on Keras v3

@odulcy-mindee All models would require a retraining

• maybe manual reading the weights from the keras v2 checkpoints -> set weights manual -> save in keras v3 format could be an alternative option to avoid retraining
• Afterwards model building isn't required

PS: the mentioned issue with sar_resnet31 is solved

I would try the way with "set weights manual" if this should not work we would need to retrain everything (but not before the next release and next year i think :) )