Committed rules documentation: rules.md, decisions.md, mechanisms.md with enforcement tests

[derek73]

Implements the library-rules-documentation design (spec developed over six evidence-grounded review rounds; process artifacts stay gitignored — this PR is the durable residue).

What this adds

Three contributor docs in docs/design/:

• rules.md — 33 normative rules of how names are written and parsed, in 12 domain-topic sections (never pipeline-stage-keyed, so refactors can't force renumbering). Each rule: linguistic rationale → implementation-free statement → executable examples → accepted consequences. Every example line runs against the live parser in CI.
• decisions.md — the ADR-style record: dated decision entries, Declined: (rejected proposals WITH their measurements), Excluded: (standing vocabulary prohibitions with per-entry reasons), Open: (issue links, never restated), and the promoted 3-0-reevaluations list.
• mechanisms.md — 17 solution patterns keyed by problem shape, each with a citable contract statement, plus a Verification-shapes section (the measurement traps and their antidotes).

Three enforcement tests (tests/v2/):

• test_rules_doc.py + rules_doc.py — parses and executes every example line (with policy/locale annotations, extras gates, warns=/raises= diagnostic forms, strict deviates: markers for known gaps — currently zero).
• test_doc_citations.py — cited IDs must exist; code-comment citations must be verbatim excerpts of the statements they cite; implemented: lists must match the modules actually citing each rule; the armed legacy check keeps §-style gitignored-spec references out of the committed tree.
• test_doc_spellings.py — a denylist of spellings that have shipped wrong; its seed entry caught itself live in usage.rst and release_log.rst on first run (fixed here).

Docstring slimming: stage docstrings keep their Consumes/Produces/Reads headers plus quoted rule citations and genuinely code-local notes; the decision narratives moved to the docs (the 57-line rule-1b block is now a 10-line citation). All 51 committed references to gitignored planning docs are retargeted.

AGENTS.md gains the maintenance rules: same-PR amendment (behavior changes amend rules.md in the same diff), the counting-claims taxonomy, release-log claim sourcing, rule-shaped issue drafting, and the read-mechanisms-first pointer.

Zero behavior change, proven three ways

• AST comparison (docstrings stripped): all 23 changed nameparser/ files are byte-for-byte identical in executable structure to master.
• Full suite: 3353 passed, mypy and ruff clean.
• Differential gate at every ledgered baseline, each unexplained: 0 — 1.4.0 (108 intentional, ledgered), previous minor (1), 2.0.0 (90).

Review process

Every extraction pass ended with an adversarial probe agent trying to falsify the statements against the live parser. The probes earned their cost: they falsified nine drafted statements (the Turkic rotation's "exactly four words", the particle join's reach, the W-section division scope, and more) — each corrected to measured reality with the falsifying input pinned as an executable example.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.51%. Comparing base (5320a42) to head (3176a5e).
⚠️ Report is 3 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #381   +/-   ##
=======================================
  Coverage   98.51%   98.51%           
=======================================
  Files          44       44           
  Lines        2895     2895           
=======================================
  Hits         2852     2852           
  Misses         43       43           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.