DOC Fix docstring defaults that do not match the signature - #34772

Reference Issues/PRs

None. These are docstring-only corrections where the documented default contradicts
the signature, which the Documentation section of the pull request checklist exempts
from needing an issue ("typos or obvious inconsistencies in the documentation").

What does this implement/fix? Explain your changes.

Nine parameters in seven files documented a default value that the function or
class signature contradicts. This PR corrects the docstrings to match the code. No
behaviour is changed.

Two of these are settled by the codebase itself:

• SpectralClustering the class already documents n_clusters : int, default=8
  (_spectral.py:407) and n_components : int, default=None ... If None, defaults to n_clusters. The spectral_clustering function in the same module disagrees with
  both. These changes make the function match the class.
• NMF documents verbose : int, default=0 (_nmf.py:1458), so do the three private helpers
  in that module (lines 453, 788, 1069). MiniBatchNMF is the only one
  documenting it as bool / False.

randomized_svd's default='warn' is left over from a deprecation cycle that ended
in 1.2; the signature has been None since then.

Introduce yourself

I am saikrishna working as a senior data engineer I have bachelors degree in computer science with data analytics specialization where I have worked on wide variety of tools like numpy, pandas, scikit-learn,... Where I have done various projects using various tools. Now I was reading the source looking for a bug to fix. So I wanted to contribute and help tool perform better.

AI usage disclosure

I used AI assistance for:

• Documentation (including examples)
• Research and understanding

This pull request includes code written with the assistance of AI.
The code has not yet been reviewed by a human.

Any other comments?

On return_train_score: BaseSearchCV defaults this to False (_search.py:1696,
_search.py:2095), while both halving classes override it to True. I have documented
the behaviour as it currently is. Whether the True override was intended is a
behaviour question and I did not want to fold it into a documentation PR; happy to open
a separate issue if that is worth looking at. If reviewers would rather not touch that
one here, it can be dropped and the other eight are unaffected.

No changelog entry: check-changelog.yml only requires one when a file matching
sklearn/.+test_.+\.py changes, and this PR touches no test files.

Checks run locally: sklearn/tests/test_docstrings.py,
sklearn/tests/test_docstring_parameters.py and
sklearn/tests/test_docstring_parameters_consistency.py (2939 passed), the test modules
for all seven touched files (826 passed), and ruff 0.12.2 check and format.