Support Keras 3 models in from_keras

[nfirle]

Summary

This PR adds support for Keras 3 in the from_keras conversion function. Keras 3 introduced several breaking changes in model serialization and internal APIs, which caused the existing conversion logic to fail. This update ensures compatibility with the new format.

Changes

• Added a new function from_keras3 that handles models created with Keras 3.
• Updated from_keras to detect the Keras version and delegate to from_keras3 when necessary.
• The from_keras3 function is designed to also work with older Keras versions, but for full backward compatibility, the original from_keras logic is preserved.
• This approach minimizes the risk of regressions and keeps the behavior consistent for users still relying on Keras 2.

[andife]

@xadupre could you have a look at it?

[sharktide]

Suggested change

	#print(model_proto)

Might be better to remove the debug statement and extra newlines

[nfirle]

Hello, I just wanted to ask whether the pull request is still under review or if there is no further interest in it. If it is still being considered, is there anything I should change to make it suitable for integration?

[sharktide]

Hello, I just wanted to ask whether the pull request is still under review or if there is no further interest in it. If it is still being considered, is there anything I should change to make it suitable for integration?

I am interested, and I think this would be a useful feature, however I feel that I am not the right reviewer for this PR

[sonarqubecloud]

Quality Gate failed

Failed conditions
36.9% Duplication on New Code (required ≤ 3%)

See analysis details on SonarQube Cloud

[andife]

With the major repository changes that landed recently, this PR is now showing a merge conflicts. If you're up for it, feel free to retry and adapt the changes for the newer version — that would be much appreciated.

[andife]

does the supported tensorflow version of tf2onnx has to be updated before?