Restore accurate C# semantic token kinds in LSP and excerpt output

[Widthdom]

Summary

C# semantic classification is wrong again across surfaces: private is emitted as a variable, a method name as a type, and CLI excerpt labels using as a variable.

Audit ID: G-27. Reproduced on v1.40.3.

Existing issue relationship

Direct regression of closed #4444, which covered C# keyword/declaration semantic-token kinds.

Expected behavior

C# keywords/modifiers, namespaces/types, methods, properties, parameters, and variables must map consistently in both decoded LSP tokens and excerpt metadata.

PR scope

Keep this to the shared C# semantic token classifier/mappers.

Implementation guide for Codex 5.6 sol xhigh

• Build focused source fixtures and decode LSP delta tokens back to absolute spans/types in tests.
• Share one classification source between excerpt and LSP writers; coordinate encoding must not choose semantic kind.
• Cover using directives, access modifiers, method declarations/calls, generics, records, and contextual keywords.
• Preserve token ordering/non-overlap and supported legend compatibility.
• Add regression tests, run full suites, and add a bilingual changelog fragment referencing Emit accurate C# semantic token types for keywords and declarations #4444.

Acceptance criteria

• The reproduced modifiers/methods/using tokens have correct kinds.
• Excerpt and decoded LSP classifications agree.
• No token range or legend-index regression occurs.

[Widthdom]

Work started on branch fix-issue4852.

[Widthdom]

Implemented in draft PR #4950.

The shared C# semantic-token classification now covers decoded LSP tokens and excerpt/inspect output, including indexed-source fallbacks and bounded reconstruction. Regression tests, bilingual documentation, and changelog.d/unreleased/4852.fixed.md are included.

Local Release validation completed with 0 failures on net8.0 and net9.0. CI is currently in progress with no failures reported at the bounded final check.