It seems entirely reasonable that there needs to be a version available that's precisely as-published, for legal (and quasi-legal) reasons, as you say - however, that's the version produced by the RFC Editor, and not the tools version (which is already non-normative, technically, due to the markup).
What I'm driving at is whether the right way to handle errata is by changing the document on tools (perhaps by diff submission). This should reduce the mechanical workload of errata handling to near-zero, and leave the judgement calls of whether to accept them as the cost.