Extend the ZH surname list past the census top 100 (#695, #657)

[derek73]

Closes #695. Also covers #657 (closed, folded into #695): the missing common surnames and the missing compounds.

What changes

locales.ZH divided an unspaced Chinese name only after one of the 2020 PRC census top-100 surnames, so 柯洁 stayed whole as family 柯洁. The single-character list is now the census rows plus the union of two openly licensed rankings' top 300:

• Taiwan's Ministry of the Interior surname ranking (2023, data.gov.tw 126774, Open Government Data License v1; attribution is in zh.py's docstring).
• Wikidata's Chinese family names ranked by bearers (CC0).

tools/zh_surnames.py fetches both and writes a generated block into nameparser/locales/zh.py, so the list is re-derived rather than transcribed. It also uses Unihan's variant table to add simplified forms, in one direction only: 余's traditional variants include 餘, which nobody bears as a surname.

Compounds stay hand-curated. 25 were added (钟离 and 段干 among them) under an admission test stated in zh.py, and married double surnames like 李林 are excluded.

柯洁      →  family '柯', given '洁'      (was family '柯洁')
钟离小凤  →  family '钟离', given '小凤'  (was family '钟')

Measured

	before	after
Chinese rows of the native-script corpus right	74/80	80/80
Japanese rows split wrongly under ZH	2/60	11/60
CLDR ZH locale view	42/96	42/96 (moved renderings wrong before and after)

The default parse moves no corpus name: the pack is opt-in, and the snapshot only gains the new example names.

Decisions taken in the session (recorded in decisions/W1 (unspaced CJK division))

• The pack is for Chinese names. Applying locales.ZH declares the data Chinese, and Japanese data has locales.JA. So a common Chinese name left whole is a failure, and more Japanese wrong splits under ZH are an accepted cost.
• One pack, cut at 300. A full-roster pack and a "split after the first character" mode were both declined. Taiwan's tail past 300 covers 0.33% of its population and holds the characters that open Japanese surnames. The corpus doesn't force 300: a cut of 212 reads all 80 rows with 7 Japanese wrong splits. 300 leaves a margin over a corpus of public figures.
• Both sources, not one. On the corpus alone, Wikidata's list matched the union with 2 fewer Japanese wrong splits. But 46 surnames reach the list only through Taiwan, including 关 and 区 (both named in The Chinese surname list misses common surnames, and splits compound ones wrongly: 钟离小凤 gives family 钟 #657), 颜, 东 and 兰.
• The compound test weighs living bearers. 張簡 stays: 9,162 bearers in the Taiwan file against 9 Wikidata people named 張 + 簡, who now read whole. 东野 is left out (no bearers in the Taiwan file, against 2 such people), and so is 成公 (成公亮 is 成 + 公亮).

Accepted costs

• A Japanese name read under ZH splits after its first character more often (安藤一郎 → family 安).
• A transcribed foreign name written without the 间隔号 splits too (安娜 → 安 + 娜).
• A listed compound wins over its first character in every name that opens with it (张简明 → 张简 + 明).
• The nickname style 阿明 now reads 阿 + 明; 阿 is a Taiwan surname at rank 282.

Also in the diff

• Five tests and a case row used 阿 as an unlisted surname; they now use 小.
• 司马相如 and 諸葛亮 now record a segmentation ambiguity, because 司 and 諸 are listed.
• Re-recorded agreement files, CJK and rules corpora, and the snapshot.
• New W1 examples, a vocabulary-exclusions block for married double surnames, the user docs' zh row, and a release-log bullet.

Review

Two design-docs review rounds and one code review ran; their findings are fixed in commits 5945de4, 6fad6ae and the last commit. The suite passes without the ja extra installed, the path CI's build jobs take. mypy, ruff and the Sphinx doctest build are clean.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.04%. Comparing base (c726650) to head (ba53232).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #730   +/-   ##
=======================================
  Coverage   99.04%   99.04%           
=======================================
  Files          46       46           
  Lines        4585     4587    +2     
=======================================
+ Hits         4541     4543    +2     
  Misses         44       44           

☔ 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.