柯洁 reads as one family name under locales.ZH: should the Chinese surname list go past the census top 100? #695

What happens today. The ZH pack segments only surnames in its list, the 2020 PRC census top 100 plus the Hundred Family Surnames compounds (nameparser/locales/zh.py). Its docstring calls this "a coverage floor, not a census roster". Any other surname leaves the name whole:

parser_for(locales.ZH).parse("柯洁")   →  given '', family '柯洁'
parser_for(locales.ZH).parse("欧阳修") →  given '修', family '欧阳'

Evidence. On 80 real unspaced Chinese names from Wikidata (#693), ZH agrees on 74. All 6 misses are surnames outside the list: 詹 (詹桓), 卫 (卫惠林), 柯 (柯洁, the Go player), 茅 (茅祖權), 华 (华楠) and 文 (文强). The CLDR locale view, which uses invented names, never showed this; its 42/96 fails for other reasons. That is a 7.5% miss rate on public figures, who skew toward common surnames, so ordinary data likely misses more.

Proposed statement. Extend the single-character list to a larger published roster (top 300 or top 500), keeping the census rank order and the existing annotations.

Open questions:

• Source. The 2020 census publication gives only the top 100. Which openly licensed roster goes further? Candidates to check: the 2019 Ministry of Public Security surname release, and Wikipedia's 中国姓氏排行 page and its sources.
• Collision cost. Rarer surnames are more often also given-name characters (文 and 华 are common given-name characters). A two-character name whose first character is a listed surname is split either way, so the real cost is a non-Chinese Han name split wrongly, such as Japanese kanji read without JA. The pack is already opt-in against that. Measure against the CLDR ja rows under ZH before and after.
• Traditional forms. Should 衛 and similar be added in both scripts, as the current list does?

Exclusions. This does not change the default (ZH stays opt-in), and does not touch compound surnames beyond the existing list.