Read junior and senior in 34 more languages; let any suffix word start the credential run

[derek73]

Closes #679.

1. Any suffix word starts the credential run, whatever its script

Since #602, a trailing suffix word marks the start of the post-nominals, so John Smith PhD Jones reads suffix PhD Jones. #602 excepted non-Latin honorifics, and the code drew that line by script. #681 then added Cyrillic and Devanagari generational words, and the script line refused those too, so Serbian read differently in its two scripts.

Measured over every non-Latin suffix word (51) in six shapes under three orders (918 parses), the exception protected no correct reading. An honorific followed by a name word read as a middle name, with no report:

Input	Before	Now
राम शर्मा साहब मोहन	middle शर्मा साहब, family मोहन	family शर्मा, suffix साहब मोहन, reports suffix-or-name
Ана Хорват старији Џонс	middle Хорват старији	suffix старији Џонс, as Ana Horvat stariji Jones reads

• Parses: 360 of 576 honorific parses and 285 of 342 generational ones move.
• One-character honorifics (씨, 様, 氏, 殿, 군, 님, 양 and जी) never move, because S2's two-letter minimum already excludes them.
• Corpus: no name moves outside rules.md's own examples.
• Code: this drops _pieces._all_latin (added in Read junior and senior in other languages as suffixes #681), S2's honorific exclusion, and Read junior and senior in other languages as suffixes #681's Accepted clause. decisions.md#S2 records the retraction.

2. 76 more junior/senior words, 34 more languages

The new words come from CLDR's generation samples. Each was checked against real trailing usage on its own language's Wikipedia, and for name collisions with tools/name_usage.py.

CLDR's word turned out to be unreliable in about a third of languages. Some entries are field-label translations, some are adjectives written before the name, and some are abbreviations nobody uses. In those languages, the word the language's Wikipedia actually writes after names ships instead (lo, ti, or, ta, si, mk).

Input	Before (1.4.0 / 2.4.2)	Now
ロバート・ダウニー・ジュニア	whole text first / family ジュニア	family ダウニー, suffix ジュニア
로버트 다우니 주니어	family / middle 주니어	suffix 주니어, peeled before the hangul family-first order applies
جورج بوش الابن	—	suffix الابن

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

• Hebrew סניור (an Israeli surname) and Japanese シニア.
• Ավագ, ជូនៀ and Ajala, which are names.
• Abbreviations shaped like initials.
• Words that come before the name: az, uz, tk, ba, ky, kk.
• CLDR forms that are label translations or unattested.
• Pack-shaped: fr fils/père (the surname Fils, 767 US Census bearers) and the Latin ml./st. (cs/sk/sl/hr). Their collisions are with other languages written in the same script, so they wait on a locale pack (Locale packs for Alexandre Dumas fils and Jiří Dienstbier ml.: generational words that collide outside their language #684).

Cyrillic мл. ships. Russian, Bulgarian and Serbian all write it after names, and it collides with nothing in Cyrillic.

Measured

• Corpus: 16 CLDR corpus renderings move (is, ar, fa, gu, kn, ml, ru, cv), each gaining its generation word as the suffix.
• CLDR agreement: cldr_agreement shows 4 suffix fields newly agreeing and no regression. 40 fields changed but still disagree: in 12 suffix fields the generation word is now the suffix but the credential after the comma isn't taken (yngri where CLDR has yngri þingmaður), and in 28 family or given-name fields the generation word left a field that is still wrong for other reasons.
• Wheel checks: the readings in the two new case rows were checked against the 1.4.0 and 2.4.2 wheels.
• Checks: the full suite, mypy, ruff and the Sphinx doctests pass.

Review

• Code review: found stale мл comments and a docstring list that omitted honorifics. Both fixed.
• Design-docs review: found that my honorific grid had keyed results by each order's first role, so one order overwrote another. The corrected figures and the recompute recipe are now in decisions.md. It also found the CLDR agreement gain overstated (field gains reported as whole renderings), a missing cv in the mover list, weak evidence for excluding シニア (kept out by Derek's call), and two test comments still describing the honorific refusal.

No release-log bullet for part 1: only input with a name word after an honorific reaches it. Part 2 widens #681's existing bullet.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.03%. Comparing base (4410d90) to head (2620ddb).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #685      +/-   ##
==========================================
- Coverage   99.03%   99.03%   -0.01%     
==========================================
  Files          46       46              
  Lines        4543     4541       -2     
==========================================
- Hits         4499     4497       -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.