Remove duplicate remove_outer_derefs and refactor mod deref

[kkysen]

This refactors mod deref, implementing most of the functions there in terms of one try_remove_outer_deref_as_ref so that we know they're all in sync. Also, try_remove_outer_deref is added so that you don't have to do separate has_outer_deref and remove_outer_deref calls.

Then I used this to de-duplicate the 1 has_outer_ref and 2 remove_outer_deref calls in the RValue::{AddressOf,Ref} match arms by combining the arms and using try_remove_outer_deref.

Then I used apply to simplify the conditional builder logic. It's quite a small but useful dependency for functional, chaining code.

The remaining code is still quite similar between the RValue::AddressOf and Rvalue::Ref match arms, so maybe we can refactor that into common code as well.

[oinoom]

I'm not in favor of combining the match arms. I do however like your original idea outlined here in order to bind source in the match arm for use in the body and would support that change.

[oinoom]

i don't want to combine these, there are more complicated cases incoming and it'll be much clearer if they're separate

[oinoom]

i don't want to combine these, there are more complicated cases incoming and it'll be much clearer if they're separate

[kkysen]

I'm not in favor of combining the match arms. I do however like your original idea outlined here in order to bind source in the match arm for use in the body and would support that change.

My original idea uses #![feature(if_let_guard, let_chains)], though. It is a lot nicer, though, I think, but I don't want to add unstable features.

[kkysen]

#![feature(let_chains)] is already merged and stabilizing in Rust 1.64 (rust-lang/rust#94927), so since we're pinned to a nightly, too, I think using that should be fine, but #![feature(if_let_guard)] is not close to stabilization (rust-lang/rust#51114).

[kkysen]

I understand there might be more complex logic coming, so I shouldn't combine the logic in the way I did, but I don't think combining match arms into one per Rvalue variant is bad. We can then do more complex branching within each Rvalue match arm rather than be limited to only guards.