Do you also want to re-export these in the root? I think most people don't use the module hierarchy from here, which is honestly kind of a mess...

When a type also impls PartialEq, then ZeroConstant also provides a blanket impl of Zero, and likewise for OneConstant/One, making it simple for stack-allocated integers to impl these traits as an alternative to Zero/One while still remaining fully compatible.

The problem is that these are not perfectly overlapping, especially with generics, like:

impl<T: Zero> Zero for Complex<T> { ... }
impl<T: ZeroConstant> ZeroConstant for Complex<T> { ... }

The first would include non-const types, but a blanket impl<T: ZeroConstant> Zero for T makes these conflict.

It's similar to Niko's old post about wanting impl<T: Copy> Clone for T:
https://smallcultfollowing.com/babysteps//blog/2016/09/24/intersection-impls/

I have removed the blanket impls, and in their place impl'd ZeroConstant for Wrapping instead.

Sidebar: would names like ZeroConst and OneConst be better?

Sidebar: would names like ZeroConst and OneConst be better?

Probably yes, or ConstZero and ConstOne. There's not a lot of precedent that I can find, but some unstable APIs in std do use an abbreviated prefix, like ConstParamTy.