Join an elided particle or Arabic article written after the given name to the family (#700)

[derek73]

Closes #700.

Library catalogs file Paul d'Ivoi as Ivoi, Paul d' and Ḥusnī az-Zaʿīm as Zaʿīm, Ḥusnī az-. That is, the particle is written apart, behind the given name, with its mark still on it. Neither written form was particle vocabulary, so 1.4.0 through 2.4.2 all read it as a middle name. Since rules.md#P6 asks only the particle tag, the parse fix is vocabulary alone.

Input	Before	After
Ivoi, Paul d'	middle d', family Ivoi	family d' Ivoi
Rashid, Harun al-	middle al-	family al- Rashid
Zaʿīm, Ḥusnī az-	middle az-	family az- Zaʿīm
Paul d' Ivoi	middle d'	family d' Ivoi
IVOI, PAUL D' .capitalized()	Paul D' Ivoi	Paul d' Ivoi
jacobus henricus van 't hoff .capitalized()	van 'T Hoff	van 't Hoff

Decisions (yours, from the session's options)

• Elided set: d', l', dell', each also with U+2019, and de’ added beside de', because nothing folds the typographic apostrophe.
• Articles: a closed list, not a shape rule. It covers al-, ul-, the article assimilated to each sun letter in both ASCII and DMG spellings, and the Egyptian el- series. All are never-given, while the bare al (Al Gore) stays ambiguous.
• Rendering: joined with a space like every particle (d' Ivoi). Glued forms like d'Ivoi are unchanged.

What review changed

• esh-, eth-, edh- added. The design-docs review found that the el- series lacked the ASCII digraphs the a- series had. They weren't in the first approved list; you kept them after review, since the trailing hyphen keeps them clear of any name word.
• Case repair fixed (_render.py). capitalized() split a token at its mark and tested the stem, which gave D' Ivoi. The whole token is now asked the particle arm's question first. That arm is now one shared predicate, _lowers_as_particle, and the early return takes only its lowercase. The second review caught a first version that also moved ed- and configured hyphenated particles.
• 't reached as a side effect. The case-repair fix also reaches the shipped Dutch 't. It now lowercases in van 't Hoff, which is R4's reading. This is a change from every release, so it has its own release bullet.
• Doc corrections. The P6 sentence no longer claims "never a given name". The Ivoi, d' boundary example came out of P6 because the rule's statement implies otherwise. The case row still pins it.

Measured

• Parse: against b4a004c3, 39 corpus names move (38 GND, 1 conventions). Each one has a trailing d' or article that leaves the middle name and joins the family, and no ambiguity kinds change. To recompute: uv run python tools/differential/snapshot.py --since b4a004c3. The GND and conventions agreement records gain that progress.
• Render (not visible in the snapshot): against 00dd2759, 28 corpus names move. Each move only lowercases a token ending in ' or -. A particle × position × order grid adds 'T → 't.

Docs

• rules.md#P6: a sentence and examples.
• rules.md#R4: three capitalized examples.
• decisions.md#P6: a dated entry with a Declined line and the not-decided scope.
• Release log: two bullets.
• AGENTS.md: a lesson on checking the render views when adding a vocabulary word that contains punctuation.

PR review round (code, tests, comments, simplifier)

• Decline path pinned. test_a_marked_particle_lowercases_only_where_the_particle_arm_would covers the all-particle part, the declared given slot, the suffix ed- and a configured van-der. The first broad early return had passed the whole suite; I measured that it fails this test.
• ’t added beside 't. particles.py now asserts at import that every elided entry has both apostrophe spellings.
• Ivoi, d' relabelled fix(#700). It had been labelled parity by default, but 1.4.0 through 2.4.2 read given d'. The same move happens to any new entry standing alone after a comma (Smith, al- → family Smith al-), as Jong, de has read since 2.0. That shape is garbage input, so it has no release bullet.
• New rows for the empty head, a particle-surname head, a glued comma, and ’t.
• Frame cost removed. The whole-token check is gated on not text.isalpha(), because it had cost every token two frames under capitalized().
• Prose corrections, including AGENTS.md, which had wrongly called initials() invisible to the snapshot.

The full suite, ruff, mypy, and the Sphinx doctests all pass, and the suite ran without namedivider, the way CI runs it.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.03%. Comparing base (b4a004c) to head (c7ee4f1).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #711   +/-   ##
=======================================
  Coverage   99.03%   99.03%           
=======================================
  Files          46       46           
  Lines        4552     4557    +5     
=======================================
+ Hits         4508     4513    +5     
  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.