This lets us clean up some little things (see this MR), and it enables some major improvements to gmap and Qp by relying on definitionally proof-irrelevant propositions.
I asked last evening on Mattermost if anyone still uses std++ with such old versions of Coq. So far, nobody responded, but it has been less than 24h. On the other hand, such users can just stick to std++ 1.4.0.
