> If you use bytes that are invalid in UTF-8 (e.g. 0xF5-0xFF) as delimiters / structure characters

No need for that. UTF-8 is a superset of ASCII, and ASCII already includes a handful of control codes that could be used here.