Summary

• DBSCAN: clarifies that n_jobs parallelizes the radius neighbor query across samples during fit, and has no effect when metric="precomputed"
• HDBSCAN: explains the algorithm-dependent behavior — brute splits pairwise distance matrix into row slices, while kd_tree/ball_tree parallelize the kNN query across samples
• OPTICS: clarifies that the kNN query for the reachability graph is parallelized across samples during fit
• MeanShift: minor cleanup removing a duplicate Glossary reference (in both the class and function docstrings)

SpectralClustering is already well-documented and left unchanged.

Contributes to #14228.

Test plan

• Verify rendered docstrings render correctly
• Confirm no test regressions in sklearn/cluster/tests/

🤖 Generated with Claude Code