Guard unsupported pipeline sample weights

[FlorianPfaff]

What changed

• Only wrap scikit-learn Pipeline estimators when their final estimator explicitly accepts sample_weight.
• Preserve the existing unweighted fallback for pipelines such as StandardScaler → KNeighborsClassifier instead of forwarding an unsupported keyword and failing during fit().
• Add a regression test alongside the existing supported-pipeline routing test.

Root cause

The runtime adapter wrapped every pipeline whenever sample_weight was supplied. Its fit() method then unconditionally forwarded <final_step>__sample_weight, even when the final estimator did not expose that parameter. This bypassed the capability check already used by the core random-subspace implementation.

Validation

• Diff is limited to the sample-weight adapter and its focused test file.
• The new test exercises an unsupported final estimator and asserts the ensemble still fits and predicts.
• Full repository validation is delegated to GitHub Actions because the local runtime does not provide the GitHub CLI or a network-accessible checkout.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ ACTION	actionlint	29		0	0	1.22s
✅ COPYPASTE	jscpd	yes		no	no	3.96s
✅ MARKDOWN	markdownlint	141	0	0	0	2.5s
✅ PYTHON	ruff	1558	0	0	0	0.62s
✅ REPOSITORY	git_diff	yes		no	no	0.14s
✅ YAML	prettier	70	0	0	0	2.3s
✅ YAML	v8r	70		0	0	8.12s
✅ YAML	yamllint	70		0	0	1.49s

Notices

📣 MegaLinter 9.5.0 is out! Discover the new features and security recommendations in the release announcement. (Skip this info by defining SECURITY_SUGGESTIONS: false)

See detailed reports in MegaLinter artifacts

Show us your support by starring ⭐ the repository

[github-actions]

Test Results

     4 files  ±0       4 suites  ±0   4m 41s ⏱️ -2s
 6 866 tests +1   6 828 ✅ +16   38 💤 ±0  0 ❌  - 15 
27 472 runs  +4  27 320 ✅ +64  152 💤 ±0  0 ❌  - 60 

Results for commit c0a7f78. ± Comparison against base commit 67e9a68.

♻️ This comment has been updated with latest results.