Patch tf2onnx to ensure compatibility with numpy>=2.0.0

[james77777778]

This PR adds hacks to tf2onnx to ensure Keras remains compatible with numpy>=2.0.0.

Additionally, the export logic has been simplified, removing the need for a temporary SavedModel.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 42.98246% with 65 lines in your changes missing coverage. Please review.

Project coverage is 81.83%. Comparing base (94977dd) to head (7de1d2b).

Files with missing lines	Patch %	Lines
keras/src/export/tf2onnx_lib.py	32.29%	59 Missing and 6 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #20725      +/-   ##
==========================================
- Coverage   81.93%   81.83%   -0.10%     
==========================================
  Files         551      552       +1     
  Lines       51264    51363      +99     
  Branches     7924     7944      +20     
==========================================
+ Hits        42003    42034      +31     
- Misses       7315     7375      +60     
- Partials     1946     1954       +8     

Flag	Coverage Δ
keras	81.65% <41.22%> (-0.10%)	⬇️
keras-jax	63.95% <40.35%> (-0.07%)	⬇️
keras-numpy	58.81% <11.40%> (-0.11%)	⬇️
keras-openvino	29.84% <11.40%> (-0.05%)	⬇️
keras-tensorflow	64.62% <40.35%> (-0.07%)	⬇️
keras-torch	63.94% <11.40%> (-0.11%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[fchollet]

LGTM, thank you!