Revert CUDA 12.8 shared workflow branch changes

[vyasr]

This PR points the shared workflow branches back to the default 25.02 branches.

xref: rapidsai/build-planning#139

[betatim]

I don't think the failure in https://github.com/rapidsai/cuml/actions/runs/13068021505/job/36466764589?pr=6282 is related to the changes in this PR. However, we also haven't merged anything in the last few days -> the test result shouldn't have changed from e.g. #6266.

This is the output:

   File "/opt/conda/envs/test/lib/python3.12/site-packages/pluggy/_manager.py", line 120, in _hookexec
    return self._inner_hookexec(hook_name, methods, kwargs, firstresult)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/test/lib/python3.12/site-packages/pluggy/_callers.py", line 139, in _multicall
    raise exception.with_traceback(exception.__traceback__)
  File "/opt/conda/envs/test/lib/python3.12/site-packages/pluggy/_callers.py", line 103, in _multicall
    res = hook_impl.function(*args)
          ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/test/lib/python3.12/site-packages/_pytest/python.py", line 194, in pytest_pyfunc_call
    result = testfunction(**testargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/__w/cuml/cuml/python/cuml/cuml/tests/test_kmeans.py", line 174, in test_weighted_kmeans
    assert cu_score - sk_score <= cluster_std * 1.5
AssertionError: assert (-2417.9619140625 - -14819.1552734375) <= (1.0 * 1.5)

At first glance both the cuml and scikit-learn estimators use a seed, as does the data generation. For scores the rule is "bigger is better", so in this case I think the cuml score is better. I remember talking about scores, cuml score's being better and CI failing because of it with @dantegd a while back. Do you remember where that was? Should we apply the same fix here?

Though it is curious that the results of the two estimators would be this different, all of a sudden.

[bdice]

/merge

[bdice]

@betatim Your observations sound right, I'll leave it to you and @dantegd to figure out what to do. It seems the test failures are in optional jobs so I hope they do not block this merge.