Mostly to provide feedback for https://github.com/coq/coq/issues/7872, but also because with this we use lia consistently everywhere. I did the same in Iris already, the speedup was negligible though.