[CI] QuantileTransformer tests fail against scikit-learn 1.10.dev0

[csadorf]

Summary

The nightly-dependencies wheel test job reports 88 assertion failures in the QuantileTransformer and quantile_transform tests when testing against scikit-learn==1.10.dev0. The failures cover NumPy, CuPy, and cuDF dense inputs, and SciPy and CuPy CSC sparse inputs.

Failing test/component: test_preprocessing.py::{test_quantile_transformer,test_quantile_transform,test_quantile_transformer_sparse}

Failure observed in:

• https://github.com/NVIDIA/cuml/actions/runs/35706866779/job/106677868890
• https://github.com/NVIDIA/cuml/actions/runs/35605487149/job/106389458129
• https://github.com/NVIDIA/cuml/actions/runs/35495938521/job/106038781196
• https://github.com/NVIDIA/cuml/actions/runs/35427823272/job/105856766940

An earlier run reported eight failures from the same test group:

• https://github.com/NVIDIA/cuml/actions/runs/34521923059/job/103021343995

Environment

• OS: Ubuntu 26.04, amd64
• Python version: 3.14.7
• CUDA version: 13.3
• GPU: NVIDIA RTX PRO 6000 Blackwell
• Dependencies: nightly-deps; scikit-learn==1.10.dev0

Test details

• Test file: python/cuml/tests/test_preprocessing.py
• Test names:
  • test_quantile_transformer (24 parameterizations)
  • test_quantile_transform (48 parameterizations)
  • test_quantile_transformer_sparse (16 parameterizations)
• Error message:

  AssertionError:
  Not equal to tolerance rtol=1e-05, atol=1e-05
  
  Mismatched elements: 600 / 600 (100%)
  Max absolute difference among violations: 0.24746089
  Max relative difference among violations: 0.79864596
  

The four most recent affected jobs each reported:

88 failed, 13638 passed, 3800 skipped, 121 xfailed, 13 xpassed