Combining characters have already made Unicode text stateful.
Although I agree that encoding length hints into it seems like a bad idea - it creates an opportunity for the encoding to disagree with the reality of the text. You need _some_ way of handling it if it says that the next grapheme cluster is 4 characters long but it's actually only three.
But even for adding it to Unicode, your proposal would make text stateful (even over long distances) which is a really bad idea.