modify the lexer to allow underscores in an identifier, but keep scanning single underscores as a specific underscore token