Drop the legacy Result serialization behavior in 5.x

[itegulov]

Something to consider for 5.x (or another major release):
1. Making #745 the default behavior for all functions that return Result
2. Deleting #[handle_result] annotation
3. Removing compilation error introduced in #748

[uint]

Following the discussion in #1057 (comment), I don't think we're going to do this. Should we close this, @frol @agostbiro?

[uint]

David Tolnay invented a way to emulate specialization that basically abuses deref coercion.

This weekend, I figured out a way to use this to emulate associated functions with trait specialization.

This looks like a promising way to reliably detect Result<_, _> types in macro expansions, including for the purposes of generating schemas. I might experiment with that sometime soon.

[frol]

@uint Oh, I missed the previous comment, sorry. Well, if we can do that and the UX won't be confusing, I am all ears!