Currently, we have just a single bucket for nodes - known_nodes.

But we need to bucket them separately. Nodes that are palpably offline don't belong in this bucket. Nor do those whose metadata is out of date. Or probably even those who have a known-but-out-of-date fleet state.