Holds diagnostic tables while their owner is down, and starts the next owner.

Every Mob.Diag.Store table names this process as its ETS heir, with the store as heir data, so a store owner that dies hands its tables — rows intact and still writable, since they are public — here instead of deleting them. The heir then starts a replacement owner, which takes them back with give_back/2. Nothing else would: Mob.Diag.Store.ensure/1 only checks that a store's tables exist, and they do, so without a restart the heir held them until it died too and took every row with it (MOB-302).

The restart runs in a process of its own. The new owner's setup calls give_back/2, which the heir could not answer while starting it, and a store that fails to start must not take the heir down. Started on first use and unlinked, like the owners.