Interesting concept! But I'm struggling to understand what makes this strictly better than just having agents push to separate git branches and then having an orchestrator agent resolve the merge conflicts? I've been doing that for a few months with auto-PRs and it seems to work mostly fine, even if it's a bit slow. Is the lockless CRDT really necessary?