fix refining_impl_trait suggestion with return_type_notation

[enthropy7]

using #![feature(return_type_notation)] on top of refining_impl_trait made the lint suggest a pretty wild “wrap the body in <Self as Trait>::f(..)” fix instead of the simple “just copy the return type from the trait”. this patch makes the lint always suggest the plain return-type replacement based on the trait signature (by grabbing the original snippet when possible), so you don’t get RTN-style desugarings in the help

new test here

fixes #151663

[rustbot]

r? @davidtwco

rustbot has assigned @davidtwco.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[fmease]

I'm interested in the reason why the type gets printed that way under RTN. I would look in the HIR pretty printer to look for the culprit. Once identified, there might be a more robust solution because using span_to_snippet is more like hot patching the bug (e.g., if the source is not available (so None) I'm pretty sure we will continue to print the type that way under RTN).

[enthropy7]

pretty_print_rpitit used RtnMode::ForDiagnostic by default. it gives impl Trait { T::method(..) } while RTN.