tree_sitter_accuracy_audit.py: csharp ground truth misses a real method due to a tree-sitter parse error in a large corpus file #1427

Summary

Found while independently verifying #1418 (csharp func_start Invocation Shield fix). After that
fix, tree_sitter_accuracy_audit.py --lang csharp's "Sample extra" output still lists
AccumulateExplicitInterfaceName from roslyn/LanguageParser.cs even though the fix reduces
GitGalaxy's own matches for that name from 4 down to exactly 1 (the real definition) -- confirmed
three independent ways: raw regex .finditer(), an isolated single-file galaxyscope scan, and a
direct SQL query against the corpus DB.

Root cause: tree_sitter_language_pack's C# grammar hits a parse error somewhere in this
14,680-line file (tree.root_node.has_error == True), and the resulting ground-truth parse tree
has no method_declaration node covering AccumulateExplicitInterfaceName at all. Since the
audit tool's "extra" classification is a straightforward set-difference against that ground-truth
set, ANY GitGalaxy match for this name -- including the single, correct one -- gets misclassified
as "extra" regardless of correctness.

This is the same class of issue as #1311/#1319 (audit-tool's own ground-truth gap, not a
GitGalaxy regex defect) -- filing per that established precedent rather than re-chasing it as a
GitGalaxy bug.

Suggested next step

Narrow down where in roslyn/LanguageParser.cs the C# grammar's parse error occurs (bisect the
file, or inspect tree.root_node.has_error/error node positions directly) to confirm the scope
of what else in this file's ground truth might be similarly undercounted -- this could mean
csharp's real extra_functions/args_exact_match numbers for this specific file are
systematically unreliable beyond just this one name, not just off by one entry.

Scope note

This is a measurement-tool defect in tests/tools/tree_sitter_accuracy_audit.py, not a
GitGalaxy structural-signature defect -- no language_standards.py/detector.py change implied.