to_additive speedup

Stream: mathlib4

[Floris van Doorn]

@Sebastian Ullrich was working on profiling mathlib, and to_additive was the slowest interpreted piece of code. I addressed these issues in !4#3632 (with the help of @Kyle Miller). This should speed up to_additive significantly, probably by 10x or more. For one of the slowest declarations for to_additive the speedup is 150x.