I was going to point out merkle-clocks, but I see that @fintohaps has already brought up merkle-crdts paper.
I also would like to point out that assumption that use CRDTs on issue comments / summary implies (near) real time text editing. While that is usually the case it’s not a requirement you e.g. in sited JSON-CRDT library you can represent issues summary as an atomic value (which is default unless you use Text CRDT) which avoids most of the mentioned complexity.
It is also worth pointing out that mentioned automerge library gives you a way to resolve conflicts differently, meaning you can additional logic e.g. you could drop edits from non-author or present them differently (maybe something like google doc suggestion edits).
I think it is also worth considering the fact that while CRDT libraries like auto-merge use specific encoding for operations it is still visible to encode that in more human readable form like in markdown files which is from what I understand is what desired here.