Add C#, C++ and C

[lightningpixel]

Builds on #6 (Java) — it is merged into this branch, so the diff includes it.

C++ and C share a grammar, a query and a resolution family. @vscode/tree-sitter-wasm ships no C grammar; the C++ one parses C without error and the tags a C file can produce are a subset. The family is shared because C++ calls into C without ceremony.

Two things the grammars forced

Inside a C++ class body the grammar names an inline member field_identifier, not identifier. The obvious pattern matched nothing, and the test caught it — expected undefined to be 'method*'. That node type occurs nowhere else, which makes the match a method by construction.

int Service::run() { … } is tagged a method directly, under run rather than Service::run. Containment cannot reach it from outside the class, and a receiver call only ever reaches a method, so tagging it a function would make it unreachable from obj.run().

Only definitions are tagged. A member declared in a class body is a field_declaration, not a function_definition, so a header declaration and its definition never become two symbols — the duplication problem simply does not arise.

C# writes each modifier as its own node rather than grouping them as Java does, and states member visibility, so it opts out of inheriting export from the enclosing type.

Measured on real projects

Project	Language	Files	Calls resolved	Refused as ambiguous
curl/curl	C	1 068	42%	2%
JamesNK/Newtonsoft.Json	C#	945	22%	57%
fmtlib/fmt	C++	77	21%	29%
google/gson	Java	264	18%	32%

C gives the densest call graph of any language in the project — 21 229 edges from 10 427 symbols, with 2% ambiguity. Free functions with project-unique names are exactly what a syntactic resolver can follow.

C#, C++ and Java send most calls through a receiver on a typed object, and x.Get(…) cannot be attributed without the type of x. Refusing is the rule working; a wrong edge would be worse. C# is the weakest of the set, refusing more than half its call sites.

The README carries this table beside the language list rather than leaving it to be discovered.

Size

Packed 703 kB → 1.4 MB, unpacked 5.8 MB → 16.3 MB. The C# and C++ grammars are 4.9 and 5.1 MB against 405 kB for Java. If that is too much for a globally installed CLI, the alternative is fetching grammars on first use — but that trades away the "nothing to compile, no network" property the install section currently promises, so it is a decision rather than an optimisation.

Verification

• npm run build, npx tsc --noEmit, npm test — 385 tests pass
• npm run benchmark — ratio 0.699, success_delta 0
• Eleven tests written first and watched fail with no language for …; the C++ inline-member one then failed a second time for a real reason, which is what uncovered field_identifier
• Indexed curl, fmt and Newtonsoft.Json end to end
• README figures re-measured after the change: 399 symbols, 1 177 edges, ~34,600 for the whole-project row