Read junior and senior in other languages as suffixes

[derek73]

Closes #661. Closes #678.

What changes

31 junior/senior words join SUFFIX_WORDS. The languages are hr/sr/bs (Latin and Cyrillic), bg, ru, uk, cs/sk, fi, lt and hi, with feminine forms beside the masculine:

Input	Before (1.4.0 through 2.4.2)	Now
Antonio Pavlović mlađi	middle Pavlović, family mlađi	family Pavlović, suffix mlađi
Агриппина Младшая	family Младшая	suffix Младшая
बलबीर सिंह सीनियर	middle सिंह, family सीनियर	family सिंह, suffix सीनियर

Each word passed two checks:

• Collisions: none (tools/name_usage.py: US Census, SSA, Wikidata).
• Usage: found written spaced after the name in real titles on its own language's Wikipedia. The issue's candidate list came partly from machine-translated data, so this check mattered.

Excluded, with reasons in decisions.md#generational-words:

• kıdemli, sóisearach/sinsearach: rank words, not generations.
• Hungarian ifj./id.: written before the name.
• Bulgarian млада: a given name.
• ml./мл./st.: stay deferred under the existing Provide constants in non-Latin scripts (Cyrillic, Greek, Arabic, Hebrew) #269 note.

The #602 run start now tests Latin script instead of ASCII. rules.md#S2 lets "an unambiguous Latin suffix word" start the trailing credential run, and the code implemented "Latin" as isascii(). Without a change, the first accented members would have split one language: Pieter Bruegel stariji Jones would start the run and Antonio Pavlović mlađi Jones would not.

• A non-ASCII word now counts as Latin when every letter's Unicode name is a LATIN one. The ASCII path is unchanged; the reference row stays at 307 frames.
• New S2 examples pin both directions. Taro Tanaka さん Jones and Гай Плиний Старший Иванов keep their family names; nothing guarded those before. Mutating the test either way fails the examples.

Accepted (rules.md#S2): the non-Latin refusal is drawn by script, because the vocabulary doesn't mark which suffix words are honorifics. So Cyrillic and Devanagari generational words don't start the run, and Serbian reads by its script (Ana Horvat stariji Jones vs Ана Хорват старији Џонс). Only a name word written after the generational word reaches this; none of CLDR's renderings has one there. Settling it is #679.

Measured

• Snapshot: six corpus names move, all CLDR hr/sr renderings carrying mlađi/млађи, each gaining the suffix. The rest of the snapshot diff is the new rules-corpus example lines.
• CLDR agreement: cldr_agreement records 14 changes, all toward CLDR's fields. The bg/hi renderings don't move because those locales' formats drop the generation field.
• Wheels: the "before" readings above were checked against the 1.4.0 and 2.4.2 wheels. Antonio Pavlović mlađi was also checked on 2.0.0 through 2.4.1.
• Gate: full suite, mypy, ruff, Sphinx doctests and README pass.

Review

• Code review: found the word count was wrong (31, not 35) and a comment overstated the collision check. Both fixed.
• Design-docs review, round 1: found the ASCII test credited to the wrong clause of S2, so The credential run tested ASCII, not Latin: Pavlović mlađi Jones and Bruegel stariji Jones read differently #678's original proposal (delete the test) would have broken S2. Hence the Latin-script fix, and The credential run tested ASCII, not Latin: Pavlović mlađi Jones and Bruegel stariji Jones read differently #678 was rewritten.
• Design-docs review, round 2: found the Serbian script split, now recorded as Accepted.

AGENTS.md's "Adding a word to a config set" paragraph now says to read the rule behind any spelling test (isascii, isalpha, script checks) before adding a set's first accented or new-script member.

Follow-ups

• "Junior" in more languages: Alexandre Dumas fils reads family fils #679: more languages, plus the honorific question.
• tools/name_usage.py fails with HTTP 431 for some Cyrillic words #680: name_usage.py fails with HTTP 431 on some Cyrillic words.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.03%. Comparing base (625bc7e) to head (222ddf9).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #681   +/-   ##
=======================================
  Coverage   99.03%   99.03%           
=======================================
  Files          46       46           
  Lines        4541     4543    +2     
=======================================
+ Hits         4497     4499    +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.