Bug: Trait resolution is unsound in the presence of bound lifetimes

[Nadrieril]

To simplify the handling of lifetime binders, our trait resolution algorithm currently erases all lifetime information, even bound lifetimes such as in for<'a> fn(&'a ()). This actually matters because types that differ in bound lifetimes are distinct for the purposes of trait solving, so we're identifying distinct types.

This is unlikely to happen in practice but technically unsound. Reproducer welcome if someone can come up with one.