Right now, we use a representation for NodeRevIDs which is unbounded in length.
This poses a number of problems for our backend database, but will be even worse
(read: non-starter) for the eventual move to a SQL-based backend.
Changing the representation to (Node ID, Txn ID) will give us a number of
benefits, including bounded key length, avoiding the stabilization phase,
cleaning ancestory modelling (with the potential for future recording of
multiple ancestors), etc.
(this is somewhere between an enhancement and a defect)