feat: add preserve_output_dtypes config [noa/issue-52] swev-id: scikit-learn__scikit-learn-25102

[casey-brooks]

Summary

• add preserve_output_dtypes to global config and wrappers
• ensure pandas outputs can retain input dtypes when untouched
• cover dtype preservation, skip scenarios, and duplicate column handling

Testing

• pytest sklearn/utils/tests/test_set_output.py -q

Issues

• Optional dtype preservation in pandas output for subset-preserving transformers #52

[casey-brooks]

Test & Lint Summary

• pytest sklearn/utils/tests/test_set_output.py -q (19 passed, 0 failed, 0 skipped)
• lint: not run (not required for this change)

[noa-lucent]

Thanks for the enhancement—most of the scaffolding looks solid. I spotted one blocking edge case: the dtype preservation check uses without , so any column containing NaNs fails the match and we silently drop back to the old behaviour. That breaks the main promise of #52 for real data with missing values. Please address the equality check (and cover it with a regression test).

[noa-lucent]

NaN preservation looks solid now—thanks for adding the regression test and tightening the equality check. I left one small nit about an unused import, otherwise this is good to go.

[noa-lucent]

[nit] Looks like is no longer used in this module—could you drop the import while you’re here?