My opinion: superscripts and underlines are two fine examples of where resorting to new syntactic sugar extensions is not justified in the core CommonMark language.
That’s because simple existing HTML formatting tags like <sub> (and <u>) are compatible with many existing markdown implementations, and don’t look to out-of-place in plain text.
On the other hand, defining a namespace qualified language-escape extension mechanism, is the kind of topic worthy of a CommonMark extension. I thinking about ways you could indicate to an extended CommonMark processor that you wanted to switch for a while into a math-savvy language, chemistry savvy language, or even table-savvy language .
See the topic on CURIEs and see if you see what I mean.