Supervises the two processes of one conversation.

The history is started before the server and the strategy is :rest_for_one, so a server crash restarts only the server and leaves the history untouched. The default restart intensity applies: a server that crashes more than 3 times in 5 seconds takes the whole pair down, which is a visible failure instead of a hidden loop. A stored conversation survives that and can be brought back with Mead.resume_session/2; one without storage is gone.