Documentation style bug

Summary

#2752 added a MathFilter that converts Sphinx/MathJax math markup to native MathML at scrape time (closes #1264). It's opt-in per scraper, and it's enabled for scikit-learn, NumPy, scikit-image, Polars and Statsmodels. The pages on devdocs.io still show raw LaTeX, though, because those documentations were scraped before that change. Each one needs to be regenerated with the new filter and uploaded by a maintainer (thor docs:upload + deploy).

This issue tracks that regeneration. Versions are the ones in the October 2026 report (#2744).

Documentations with the filter enabled, to regenerate:

• scikit-learn 1.9.0 → 1.9.1 (the documentation from latex won't display in docs like scikit learn #1264 and Add math rendering library #1442; the user guide was verified locally in fix: render math as native MathML at scrape time #2752 with 2,082 equations and no fallbacks)
• NumPy 2.5 → 2.5.3
• Statsmodels 0.9.0 → 0.15.0
• scikit-image 0.26.0
• Polars 1.44.1 → 1.44.2

Documentations that also contain math but don't enable the filter yet (adding math to their html_filters and checking the output would be a separate PR each):

• TensorFlow 2.21.0 (Math rendering for TensorFlow #750)
• Matplotlib 3.10.8 → 3.11.2
• pandas 3.0.8

PyTorch ships its own pre-rendered MathML and isn't affected.

I'll send a version-update PR per documentation, following section B of the PR template, starting with scikit-learn. Maintainers: if you'd rather regenerate some of these directly, please tick them off here.

Actual style

https://devdocs.io/scikit_learn/modules/lda_qda shows \(x \in \mathbb{R}^d\) style TeX as plain text (see the screenshots in #1264).

Expected style

Rendered MathML, as on https://devdocs.io/pytorch/generated/torch.nn.batchnorm3d and in the before/after screenshots in #2752.