Serve the docs gallery from web-sized images so Google can fetch it

[patrickchugh]

Type of Change

• Bug Fix
• New Feature
• Refactor
• Documentation

What and why

Google Search Console's "Test live URL" and "Request indexing" both failed on /gallery/ with "Something went wrong", while lighter pages on the same site passed. Measuring the page confirmed it was very heavy: it embedded twelve full-size PNGs straight from raw.githubusercontent.com.

	Images on the gallery page	Total weight
Before	12 PNGs, 5,216 to 11,488 px wide	13.8 MB
After	12 WebPs, 1,400 px wide, served from the docs site	0.7 MB

Changes:

• scripts/make_gallery_images.py reads every examples/graphs/*.png, trims the surrounding white space, scales it to at most 1,400 px wide and writes a WebP (quality 85) to docs/assets/gallery/<name>.webp. It records each source PNG's SHA-256 in docs/assets/gallery/sources.json, so it is safe to rerun and skips files that are already up to date without relying on file modification times.
• The generated images are committed (15 files, 992 kB in total).
• docs/gallery.md shows the web-sized image for each example with explicit width and height, loading=lazy on all but the first image, and the no-lightbox skip class from the glightbox configuration in mkdocs.yml. Each image is wrapped in a link to the full-size PNG, so a click still opens the original. Every heading, anchor id, prompt, command and download link is unchanged; the diff touches only the twelve image lines.
• examples/graphs/README.md gains one line telling contributors to run the script after adding an example.
• The gallery strip on docs/index.md was checked and left alone: its six thumbnails are 67 to 98 kB each, all under the 150 kB threshold.

Verification:

• mkdocs build passes with no warnings, before and after.
• All twelve /gallery/#... anchors linked from the README resolve in the built page.
• The built page wraps none of the new images in a glightbox link.

Checklist

All Submissions:

• Have you checked to ensure there aren't other open Pull Requests for the same update/change?
• Have you written Documentation/Tests?
• Have you done your own code-review?
• Have you disclosed any use of AI tools and models with their version?

AI Assistance Declaration

• Tools used: Claude Code
• Model: Claude Fable 5.1
• Scope: Wrote scripts/make_gallery_images.py, generated the WebP images, rewrote the twelve image lines in docs/gallery.md, added the README line, and verified the build. Reviewed by the author.

Checklist for Changes to Core Features:

• Have you discussed any major revamp with a reviewer/maintainer first? (It's okay to just raise a PR directly for minor bugfixes)
• Have you ensured your PR is focused on one major improvement and is not trying to do too many changes at once?
• Have you added an explanation of what your changes do and why you'd like us to include them?
• Have you written new tests for your core changes, as applicable, and made sure the new tests PASS? (docs and a build script only; no core code changed)
• Have you successfully run all previous system wide tests with your changes locally? (mkdocs build; core tests are unaffected)

🤖 Generated with Claude Code