A formal spec would be nice, I agree.  But I think it is interesting that Core's inflation bug wasn't due to nuance in the consensus rules.  Having a formal spec wouldn't have helped in this case.  The bug literally allowed coins to be created out of thin air.  Something that _obviously_ was not supposed to happen.  You might call it the most important consensus rule in Bitcoin!

As to your second point about the consensus layer not changing much, I think this is tricky in practice too. For example, a hot area of research in blockchain scaling is how to parallelize block validation.  This type of work will almost certain involve refactoring consensus critical code.  If we want to eventually processes thousands, or hundreds of thousands, of transactions per second, we will need massive parallelization.