Fix TPU test broken in master.

[hertschuh]

One test is broken on TPU, the result of transforming an image has one pixel that's off.

[gemini-code-assist]

Code Review

This pull request adds a fixed random seed to the channels_first test block in test_affine_transform. Feedback indicates that this is a brittle approach to fixing numerical failures as it masks underlying issues and reduces test coverage by resetting the RNG state to values already used earlier in the test. A more robust solution would be to adjust the tpu_atol tolerance for the specific assertion rather than forcing a specific random state.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.28%. Comparing base (28a83c5) to head (26d5cbc).
⚠️ Report is 6 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #22605      +/-   ##
==========================================
- Coverage   83.28%   83.28%   -0.01%     
==========================================
  Files         596      596              
  Lines       68089    68102      +13     
  Branches    10607    10610       +3     
==========================================
+ Hits        56711    56720       +9     
- Misses       8634     8636       +2     
- Partials     2744     2746       +2     

Flag	Coverage Δ
keras	83.09% <ø> (-0.01%)	⬇️
keras-jax	59.66% <ø> (-0.01%)	⬇️
keras-numpy	55.33% <ø> (-0.02%)	⬇️
keras-openvino	53.38% <ø> (+0.04%)	⬆️
keras-tensorflow	61.03% <ø> (-0.01%)	⬇️
keras-torch	59.84% <ø> (-0.01%)	⬇️

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.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.