codegen: address style audit findings — clones, wildcards, unwrap

[iainmcgin]

Three changes from a code-style audit of hand-written sources, focused on iteration/ownership, error handling, and dispatch structure.

Changes

protoc-gen-buffa/src/main.rs — build the CodeGeneratorResponse file list by consuming generated with .into_iter() instead of cloning every name and content string. The vector is the sole owner and is not used afterward, so the clones were pure overhead on every plugin invocation (2N string clones per protoc run).

buffa-codegen/src/oneof.rs::to_snake_case — replace c.to_lowercase().next().unwrap() with result.extend(c.to_lowercase()). Removes the unwrap and correctly preserves multi-codepoint lowercase mappings (e.g. İ → i\u{307}) instead of silently truncating them. Behavior is unchanged for ASCII, which is the only thing protobuf identifiers contain.

buffa-codegen/src/{impl_message,impl_text,view}.rs — replace nine _ => unreachable!() arms over the protobuf Type enum with explicit enumeration of the unhandled variants. Each match is now exhaustive over Type's 18 variants. A future protobuf edition that adds a Type variant will produce a compile error in buffa-codegen rather than a runtime panic during codegen, which is consistent with how wire_type_byte and wire_type_token already fail. Two of the replaced arms also had no diagnostic message at all (view.rs:1021, impl_text.rs:716).

Notes

No change to generated output — the unreachable!() arms never produce tokens. Verified with task gen-wkt-types (no drift).

task lint and task test both pass.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}