feat(site): generate a static website from the dataset

[deadmanoz]

Summary

#6 asks for a static GitHub Pages site, so the blocks can be linked to and found by searching their hash. ci/generate-website.py builds one from the repository into site/: a page per record at block/{hash}/, with the hash in its URL, title and heading; an index of the 143 records, sortable by column and filterable by text, rule or kind of evidence; the rendered docs/notes.md; the reported blocks; and a sitemap.

Each block page shows the rule, the Core reject string, the Core check and what CI checks for it, taken from the two rule tables in docs/schema.md rather than written out a second time, then the decoded header, context, observations and the evidence on file: a full block, a P2SH spend proof, a coinbase proof or the header alone. It quotes the opening paragraph of the incident note whose heading lists the block's height, or lists a block with the same failing_prevout. That gives the 85 proof-only P2SH blocks, which all spend the same output, the 2012 P2SH note; a block that only shares a rule with a noted block gets no note. Relative links in the notes resolve to block pages, the notes page or GitHub.

The sanity-check workflow now builds the site on every run, so a pull request that breaks it fails before merge. Each run uploads the build with actions/upload-pages-artifact, so a pull request's run also carries the built site. On main, a deploy-website job that needs the validation job publishes it with actions/deploy-pages, so nothing reaches the site without passing validation, and runs on a branch are serialized so an older run cannot deploy over a newer one. The deploy steps match stale-blocks', which runs them in a separate workflow. Markdown==3.10.3 is pinned next to python-bitcoinlib. The pages load one local stylesheet and one small script, which follows the visitor's light or dark preference (with a toggle in the nav) and filters and sorts the index, and no external assets.

The site is completely vibe-coded, but it looks good enough. It only reads the dataset and the docs; the data, the validator and its checks are unchanged.

This needs Settings → Pages → Source set to GitHub Actions.

Closes #6.

Testing

python ci/generate-website.py writes 146 pages (143 blocks, the index, notes and reported) and a 146-URL sitemap, and two runs produce identical output. python ci/sanity-check.py --fetch-prevouts reports the unchanged 143 blocks, 16 block files and 124 proof files. All 49 tests pass, including three new ones: note links for a block named in a note heading (783426), one sharing a named block's failing spend (174012), one sharing only a named block's rule (367047) and one with none (331673); link rewriting; and escaping of record text on a block page and the index. A crawl of the generated site resolved all 1,593 internal links and note anchors, and every link to a repository file points at a file on this branch. I checked the pages in a browser at desktop and 375 px widths, in light and dark, with no horizontal scrolling. The deploy job has not run yet; it runs only on main.

[deadmanoz]

Making some basic usability improvements & removing some duplicated data presentation first, will reopen when done

[deadmanoz]

Happy with how this looks locally now, review whenever