Fix Pin::set bounds regression

[coolreader18]

Fixes #129601

Fixes the regression from #129449, where changing the bounds of the impl block containing Pin::set changed the method resolution behavior.

struct A;
impl A {
    fn set(&self) {}
}

let a: Pin<&A>;
a.set();
// before:
// - checks <impl<Ptr: DerefMut> Pin<Ptr>>::set(): `&A` doesn't impl `DerefMut`
// - autorefs -> &A: resolves to A::set()
// now:
// - checks <impl<Ptr: Deref> Pin<Ptr>>::set(): `&A` impls `Deref`! resolves to Pin::set()
// - check method bounds: `&A` doesn't impl DerefMut: error

r? @dtolnay

[coolreader18]

This is kinda subtle - I guess theoretically, this could be deemed an acceptable breaking change, but I don't think it's worth it given it's already demonstrated breakage.

[coolreader18]

Theoretically the same could happen with as_mut(), though it's much less likely that there would be a method named as_mut that takes &self

[coolreader18]

Moved all methods back to their original block, just to be safe. I still kept the order of the methods, which means that new_unchecked and into_inner_unchecked are in separate impl blocks that both have Ptr: Deref bounds.

[dtolnay]

Thanks!

[dtolnay]

@bors r+

[bors]

📌 Commit 0d6c915 has been approved by dtolnay

It is now in the queue for this repository.