disk usage metrics. This will run in a sliver, so need to assign the
metrics to the underlying pnode, and it would be better right now if
we run it only in a single sliver (the management sliver) on each node
until I decide the best approach, since it overlaps with isalive,
which does run from each sliver.