This is tentatively fixed. I'll keep an eye on it.

I've made a real instance for admin, with a real Almanac service. I short-circuited the upgrade, sync, and restart steps for this instance specifically in the daemons, since it seems like bad news to have the daemons restart themselves.

Other than not actually doing upgrade, sync, or restart, the instance behaves normally, so the normal backup triggers should fire on it (we'll also get billed for it). It then uses the real service to identify the database host and perform backups.

At least for on-demand backups, this works properly. I'll watch it over the next few days and make sure automatic backups work, too.

This introduces a small amount of magic for the "admin" instance, notably it exists but has buttons in the UI that don't do anything. I might add some UI hints to it to make this more clear.

This also creates a small amount of peril: scripts that operate on instances need to consider that "admin" has a real instance, too, and may be passed. But they generally already need to do this, and it shouldn't matter in most cases, so I think this is not too perilous. I also think it's desirable for instances to generally have as much of the same behavior as possible, and this overall reduces the amount of special casing for the "admin" instance and gives it greater access to shared infrastructure.