Unspaced Chinese/Korean names: surnames constants + longest-match segmentation #271

Unspaced CJK names ("毛泽东", "김민준") can't be parsed today because there's nothing to split into pieces. For Chinese and Korean, segmentation is cheap and fits the library's constants-driven style, because surnames come from small closed sets while given names are open-ended — only a surname dictionary is needed:

• Korean: 286 surnames total (2015/2019 census); Kim/Lee/Park alone cover ~45% of the population; about a dozen two-syllable surnames. Complete coverage in ~300 entries.
• Chinese: top 100 surnames cover ~86% of the population, ~6,000 in current use; surnames are 1 character, occasionally 2 (compound: 欧阳, 司马, …). Excellent coverage in ~500 entries.

Proposal:

1. A new surnames constants set (same SetManager machinery as titles/prefixes), shipping curated Chinese and Korean surname lists.
2. A pre-processing step (opt-in, like all localization) that longest-matches a surname off the front of an unspaced string and inserts the space — compound/two-syllable surnames matched before single ones (欧阳明 must split 欧阳+明, not 欧+阳明, even though 欧 is itself a surname).
3. Everything downstream is existing machinery: the segmenter only inserts the space; name_order=('last', 'first') (Support family-first name order via an order-spec config (name_order) #270) does the assignment.

Opt-in is required for a second reason here: a pure-Han string cannot indicate whether it's Chinese or Japanese (林 is Lin or Hayashi), so the caller must select the language. Japanese itself is explicitly out of scope — ~138,500 surnames with genuinely ambiguous split boundaries — and is covered by #272.

Depends on the name_order order-spec (#270). Would resolve the unspaced cases behind #83 and (with ('last','middle','first')) #146-style Vietnamese input; supersedes the analysis in #34.