Sweep the non-.rst prose for staleness

[derek73]

Summary

The last item on the 2.1 docs list: the places prose goes stale that grep over .rst misses. Four sites, two kinds of drift.

The Policy spellings didn't reach the docstrings

#334 moved the guide to the spellings that type-check, but only the hand-written .rst. Four _policy.py docstrings still taught the old ones, and autodoc renders them into modules.html — so the API reference and the guide disagreed on the same page.

site	was	now
_policy.py:566	Policy(maiden_delimiters={("(", ")")})	frozenset({("(", ")")})
_policy.py:587	script_orders={}	script_orders=()
_policy.py:605	segment_scripts=()	segment_scripts=frozenset()
_policy.py:625	maiden_delimiters={("(", ")")}	frozenset({...})

Verified: grep for the old spellings in the rebuilt modules.html now returns 0.

The runtime warning was handing users an arg-type error

Sharper than the docstrings, because it reaches users who never open the docs. #337's new warning ends:

…or deactivate with Policy(segment_scripts=()).

A user pastes that verbatim, and in a package that ships py.typed it fails mypy. Now Policy(segment_scripts=frozenset()).

Nothing pinned this. The warning tests match on r"ja_segmenter" and never checked the deactivation hint, which is the actionable half of the message. Added a test asserting the offered spelling is the one that type-checks, and that the old one is absent.

Two stage headers understated their inputs

Checked mechanically — every stage's declared Reads: against the policy.* / lexicon.* attributes it actually touches — rather than by eye.

• _extract declared only the two delimiter policy fields while reading suffix_words, suffix_acronyms and suffix_acronyms_ambiguous through _suffix_shaped. Not a detail: that is the mechanism letting a clause's content overrule the delimiter's verdict ((MBA) is not a nickname), and it is the one Parenthesized (née Jones) parses as a nickname where the bare née Jones gives maiden #335 would extend.
• _post_rules omitted Policy.middle_as_family.

_script_segment, _tokenize, _assign, _group and _classify were already accurate. _vocab has no Reads: line by design, being a helper whose predicates take vocabulary as explicit parameters; its "no state" claim still holds.

Verification

• uv run pytest -q — 3065 passed (was 3064; one new test)
• uv run mypy, uv run ruff check — clean
• sphinx-build -b html — exit 0, 0 warnings
• sphinx-build -b doctest — exit 0
• The corrected advice type-checks: a file containing all three new spellings passes mypy

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.48%. Comparing base (3c45594) to head (b6d71b2).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #339   +/-   ##
=======================================
  Coverage   98.48%   98.48%           
=======================================
  Files          41       41           
  Lines        2845     2845           
=======================================
  Hits         2802     2802           
  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.