Close the two remaining review findings: one real, one demonstrably not

[Reefact]

Summary

The last review pass left two open questions before 1.0.0-rc.1. One turned out to be a real
false negative in the layout checker and is fixed; the other is a gap that cannot be reached from
C# and is written down rather than coded around.

Type of change

• Bug fix
• New feature
• Breaking change to the public API
• Refactoring
• Analyzer / diagnostic change
• Tests
• Documentation
• Build / CI / tooling

Changes

• tools/style/lint-layout.sh: closesAttribute() asked whether the raw line held )] anywhere.
  A Justification quoting code can hold one — "the analyzer reads GetValues(type)] wrongly" —
  and on the line that opens a wrapped suppression that made the site look like one already
  written on a single line. Neither mode had anything to say about it: check reported nothing,
  --fix rewrote nothing, both exited 0 — which is what a clean tree answers. The bracket tests now
  read the line with the interior of its double-quoted strings removed.
• Deliberately the line is not also cut at a //. That reads a Justification citing a URL as
  a comment, takes the real )] away with it, and turns a suppression already on one line into a
  wrapped one — whose forward scan is what deleted two members the last time it was let loose. A
  )] inside a comment therefore stays read as a close, and joins the shapes the checker admits it
  cannot read: joining would fold the comment over the rest of the line, so there is no rewrite to
  offer there anyway.
• tools/style/lint-layout.test.sh: two fixtures, one per direction. A wrapped suppression carrying
  a bracket in its Justification, now reported and joined; and one already on one line carrying a
  URL, still untouched with the member below it intact — the assertion that keeps the comment from
  being cut first.
• CLAUDE.md: records which of the two readings the checker takes and why the other is worse.
• EnumMemberNameSchemaTransformer: a remark on EscapeIgnoringCase saying why walking char by
  char is exact for every name that reaches it, since it is narrower than OrdinalIgnoreCase in
  general. OrdinalIgnoreCase folds a surrogate pair as one Unicode scalar, so it answers U+10400
  and U+10428 equal where char.ToUpperInvariant applied to each code unit says they differ; 260
  non-BMP scalars are cased that way. None can arrive: what reaches the method is an unannotated
  member's C# name, and C# has no identifier carrying a supplementary-plane character. All three
  spellings were put to the compiler — the character itself, \U00010400, and the surrogate pair as
  two \uXXXX escapes — and each is CS1056. A declared name may hold one, but it is matched
  ordinally and written by EscapeForJsonSchema, the same walk on both sides; and it cannot arrive
  by the other door either, since reaching EscapeIgnoringCase means also being the C# name of an
  unannotated member, which EMN0005 refuses before a schema is built. No code changed.

Testing

• dotnet build -c Release — clean, warnings are errors here
• dotnet test -c Release — 972 passed, 0 failed
• tests/PackageSmokeTest/run.sh — run on this branch, green; not strictly required, since
  packaging, the analyzers and the public entry points are untouched
• tools/style/lint-layout.test.sh, then tools/style/lint-layout.sh over the repository
• tools/release-notes/changelog-section.test.sh and tools/commit-lint/lint-commit-message.test.sh

Beyond the two findings, three further candidates were reproduced and then rejected, so they are
recorded here rather than acted on:

• A differential harness over six contract enums (kebab names, accents, ß, prefix-overlapping
  names, regex syntax, one non-[Flags]) and roughly 11 000 inputs — casings, commas, trailing
  commas, and the twenty-five whitespace code points — found no divergence between the binder,
  System.Text.Json and the emitted OpenAPI pattern.
• List<TFlags?>: the document types the component string while the server accepts a null
  element. ASP.NET Core emits {"type": "string"} identically for List<X?> and List<X>, so the
  nullability signal is absent rather than dropped, and inventing one would over-promise on the
  non-nullable use. Present with and without this package.
• The nullable fix on a component serving both a nullable and a non-nullable use: stock ASP.NET Core
  already emits enum: [..., null] there, and the transformer produces the same plus
  type: ["null", "string"]. The server answers 200/400 identically either way, so the fix
  reproduces what the platform says and adds nothing.

Public API

• No change to the public surface
• The surface changed and the baseline was updated in the same commit

Documentation

• README / docs/ updated
• The French counterpart was updated to match
• CHANGELOG.md and docs/for-users/CHANGELOG.fr.md both updated
• No documentation change required

Nothing under docs/ changed: the two documentation edits are a <remarks> block and CLAUDE.md,
neither of which is a paired page. No changelog entry either — neither change is visible to a
consumer of the packages.

Unrelated to this branch, but on the path to tagging: CHANGELOG.md still carries only
## [Unreleased], so tools/release-notes/changelog-section.sh 1.0.0-rc.1 exits 1 and the release
workflow would stop at "Resolve the release notes" — before any build. The section has to be written
under the exact name the tag will carry.

---

Generated by Claude Code