We used to ignore all other mounts when the first one is for the root
filesystem. But filesystems can have more than one mount and we should
look at each one individually.

This also prepares the code for more cases.

The subvolume polling is now done by a long-running monitor program
that only reports changes back to Cockpit. This reduces traffic in
general and allows us to do additional clever things.

The monitor program can optionally keep all btrfs filesystems mounted
in a "secret" location, so that we can list the subvolumes of all
btrfs filesystems, even those that are not officially mounted
otherwise.

This mode is only enabled in Anaconda mode: It is important there
since filesystems are normally not mounted when preparing storage for
Anaconda, and outside of Anaconda mode, keeping long standing secret
mounts is deemed to invasive.

The program can also carry out btrfs operations (such as creating
subvolumes) while temporarily mounting it. This allows Cockpit to
always create and delete subvolumes. The UI still only does it when
monitoring subvolumes works as well, however.