Move mc and ste to never-given, add the Spanish/Portuguese articles (#360)

[derek73]

Closes #360.

Membership: never-given 28 → 33. mc and ste move out of the
ambiguous half; los, las and das are added, having been absent
from the vocabulary entirely.

Mc Donald         given='Mc'      →  family='Mc Donald'
Ste Marie         given='Ste'     →  family='Ste Marie'
Maria das Neves   family='Neves'  →  family='das Neves'
Anjali Das        family='Das'       unchanged  ← negative control

C-i's positional qualifier is what makes das safe

Das is a borne Bengali surname — but a trailing one, and the
leading-particle rule never reaches that position. So never-given das
leaves Anjali Das and Bimal Das alone while giving Maria das Neves
back the particle it was losing. Under C-i's first draft, which asked
only whether a bearer exists anywhere, this fix would have been
declined.

What #360 actually yielded

Its premise is that the split was never curated, so many members are
probably misfiled. Applying C-i with the corpora as evidence: the
conservative default is correct for 37 of 39.

• 12 members are attested leading in the corpora and now carry that
  evidence: Al Gore, Della Reese, Mac Miller, La Shawn, San San Chiou,
  Te Awanui-a-Rangi, Van Johnson, Abu Bakr, Del Toro, Dela Cruz, Di
  Caprio, von Braun. Plus le/du/bin/do on Vietnamese and
  Chinese surnames and Provide constants in non-Latin scripts (Cyrillic, Greek, Arabic, Hebrew) #269's recorded judgment.
• von is the instructive one. von Braun is misparsed today and
  it still stays ambiguous, because Von Miller is a real bearer in
  exactly the acting position. A word can look like a pure particle, be
  misparsed as one, and still have a bearer.
• dela stays — Dela is a short form of Adela and an African name
  meaning saviour/redeemer, so the Filipino surname particle isn't the
  only reading.
• 15 stay bare: aan, aen, bat, bon, da, dal, den, dí,
  heer, santa, tho, thoe, vande, vander, vel. No evidence
  either way, and C-i's default says leave them. A bare entry means
  unexamined, so the remaining surface stays visible without a ledger —
  recording "no evidence found" would log ignorance as a judgment.

Verification, and a caveat that matters

The differential exits 0 at all three baselines and that is not
evidence here. Zero of 782 corpus names contain mc, ste, los,
las or das — not even Mc Donald or Ste Marie, the names this
issue is about. The population the change could have moved is empty, so
#360's request for "its own differential run so the numbers attribute to
a cause" has no numbers to attribute.

The proof is tests, and there were none before this commit:

• the lexicon-derived class sweep picks the five new members up for free
  — it reads particles - particles_ambiguous, so membership edits
  extend it automatically
• a new parametrized case carries the multi-word shapes plus the
  trailing-surname negative control

Both run against Lexicon.default(), not the module's reduced _LEX —
a fixture omitting these words would pass while proving nothing, which
is exactly the trap that made a #391 test inert. Mutation-checked:
reverting mc fails the new case.

rules.md#P1 gains Mc Donald and de los Santos as executable
examples. 3468 tests, ruff and mypy clean.

Follow-up worth filing

corpus_issues.jsonl is generated from the tracker and predates #360,
so regenerating it would make this whole class visible to the harness.
That's a corpus refresh across every newer issue, not part of this
change.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.51%. Comparing base (2023af1) to head (6389f06).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #393   +/-   ##
=======================================
  Coverage   98.51%   98.51%           
=======================================
  Files          44       44           
  Lines        2895     2895           
=======================================
  Hits         2852     2852           
  Misses         43       43           

☔ 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.