fix: preserve UTF-8 across stdout chunks

[dreamyt]

Summary

• decode Codex app-server stdout with StringDecoder so UTF-8 characters survive arbitrary Node stream chunk boundaries
• add a regression test that splits a three-byte Chinese character across three chunks

Problem

createJSONRPCReader called Buffer#toString() independently for each stdout chunk. If a multibyte UTF-8 character crossed a chunk boundary, Node emitted replacement characters. The resulting JSON remained parseable but its text was corrupted, which could also defeat downstream content-based history deduplication and surface duplicate messages.

Testing

• npx vitest run src/__tests__/StdUtils.test.ts
• npm run typecheck
• npm test
• npm run build