I think this is an implied goal of #905, the actual nodes.conf contents would not have any data structure associated with them and be kept in sync between memory and disk, with any individual nodeinfo things being pulled out of that in a read-only fashion dynamically as-needed.

A side effect here is getting the code path for merging nodes with assigned nodeprofiles into a separate place where it can be improved to do a better job of merging these. I think with this model it only has to merge into a read-only structure (which could still be a map without any formal data structure to it), so that merging logic can be as complex as needed. Any commands that update the config would work directly in the flatter/unmerged in-memory config then be flushed to disk.