Protocol Buffers (.proto) files are not extracted #4510

Problem

.proto files are not in CODE_EXTENSIONS, so Protocol Buffers schemas are skipped silently. In a repo that defines its service contracts in protobuf (gRPC services, or message schemas shared over a message bus), the contracts are absent from the graph:

• explain "OrderRequest" finds nothing, and there is no node that a producer's or consumer's code could be traced through

Proposed behavior

Extract .proto files (proto2, proto3, editions) as code:

• Nodes: the file, messages (nested), proto2 groups, enums and their values, services, rpcs and extension fields. The fully qualified protobuf name goes in metadata.fqn.
• Edges: contains / method for structure; references for field, map, oneof, group and extend types and rpc request/response types; references from a declaration to the extension behind each custom (option) it sets; imports for import / import public.
• Doc comments: leading doc comments become rationale nodes, attached the way protoc's SourceCodeInfo attaches them.
• Metadata: deprecated, enum value numbers, rpc streaming and google.api.http bindings. On the file node, syntax / edition, package, and the code-generation options (java_package, go_package, csharp_namespace, ...).
• Resolution: names resolve with protobuf's own scoping rule, in-file during extraction and across files in a language resolver. As with protoc, only imported files are in scope, and ambiguous references stay unbound rather than guessed.

Alternatives considered

• There is no tree-sitter grammar for protobuf on PyPI.
• proto-schema-parser (pure Python, ANTLR) parses well, but its AST carries no line numbers, so nodes would lose source_location.
• A small dependency-free tokenizer over the declaring parts of the language keeps exact line numbers and needs no new extra.

Compatibility

Additive: .proto files were not extracted before, so no existing node ids, cache entries or graphs change.

Implementation: #4509