[BUG] Scipy 1.15 Affects numeric performance of ARIMA pytest

[dantegd]

The case test_arima.py::test_integration[float64-key9-data9] after the updates in #6207 gets a significant difference in value:

AssertionError: assert np.float64(4.573729490839957) < (np.float64(1.186664497564694) * (1.0 + 0.01))
 +  where 0.02 = <test_arima.ARIMAData object at 0x7f513f06fef0>.tolerance_integration
================================================================================================================================= warnings summary ==================================================================================================================================
../testing/utils.py:366
  /home/coder/cuml/python/cuml/cuml/testing/utils.py:366: PytestUnknownMarkWarning: Unknown pytest.mark.stress - is this a typo?  You can register custom marks to avoid this warning - for details, see https://docs.pytest.org/en/stable/how-to/mark.html
    return pytest.param(*args, **kwargs, marks=pytest.mark.stress)

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
============================================================================================================================== short test summary info ==============================================================================================================================
FAILED test_arima.py::test_integration[float64-key9-data9] - assert np.float64(4.573729490839957) < (np.float64(1.186664497564694) * (1.0 + 0.01))