Tighten the docs, and stop testing them editorially

[jbloom]

Follows the hand-editing of docs/ to cut implementation detail. Checking the shortened prose against the code turned up several places where it had changed a claim into a false one.

Accuracy fixes

• The CSV's representation column is additive. It had been rewritten as "how residue is drawn", but annotation_rows sets extra_rep alongside base_rep, and docs/spec.md still said "added on top" — the two pages contradicted each other. That additivity is what produces the cartoon-plus-sticks figure.
• Unlisted heteroatoms do not take default_color. The het_layer components deliberately get no color node, so an unlisted ligand, ion or glycan keeps Mol*'s element coloring or its SNFG symbol.
• Two wrong counts in docs/examples.md. d-3-1-to-d-3-1-1.csv was called "12 sites"; it is 12 rows over 4 sites, one row per protomer. Same error in the row below it. Verified by counting unique residues.
• docs/index.md said residues are styled from a YAML file. They are styled from the CSV; the YAML is the spec that names it. README.md already said CSV.
• The Labels checkbox appears when a label was actually drawn, not whenever the CSV asks for one — a row naming a residue the view excludes draws nothing and is reported as a warning.
• assembly was marked au "(default)" although it is a required top-level key, and the orientation section no longer said which of its four keys are required (position/target are; up/radius are not).

Plus a broken bold marker (**…*) in the H1 example intro, the typos (HTmL, esidue, hiddien, "are shown are hidden", "lastest"), and a rewrap to the ~90 columns the rest of docs/ uses. Reset view gets back the one clause saying why it exists.

Anchor validation moves to mkdocs

validation.links.anchors: warn in mkdocs.yml replaces test_cross_page_anchors_resolve and the _heading_anchors helper. MkDocs resolves anchors from the rendered HTML, so it also covers the mkdocstrings headings on python-api.md that the hand-rolled slugify could not see. Verified it fails the build rather than passing silently:

WARNING - Doc file 'viewer.md' contains a link 'internals.md#no-such-heading',
          but the doc 'internals.md' does not contain an anchor '#no-such-heading'.
Aborted with 1 warnings in strict mode!

The check moves from the pytest job to the docs job; docs.yml already builds on pull_request, so CI coverage is unchanged.

That also removes a "negative control" asserting docs/spec.md contained a specific internals.md#… link. It pinned an editorial choice — cutting one sentence failed a test with nothing actually wrong — and it did not even exercise the regex it claimed to guard, matching a different pattern from the sweep's. The same content-pinning assertion (>= 10 repo links in examples.md) is dropped from the repo-link control.

tests/test_docs.py: 8 tests → 5. The three that catch genuinely silent breakage nothing else enforces (Mol\* escaping, blob/main/ links, OPTION_DOCS → docs/spec.md) all stay. CLAUDE.md gains the rule this follows.

Verification

scripts/check.sh — 215 passed. scripts/build_docs.sh — mkdocs build --strict clean with anchor validation on.

No changes to config.yml or data/; no renderer or template changes, so nothing about what gets rendered changes.

🤖 Generated with Claude Code