Generalize io for Arc

[CAD97]

There's no fundamental reason to limit the io trait implementations on Arc to only File; they can equally be applied to any Arc<T> where &T: Trait. Arc isn't a fundamental type, so it's impossible for downstream crates to implement Read for Arc<LocalType>, thus this blanket impl is a non-breaking one (if I remembered the API evolution rules correctly).

ATTN: this PR makes insta-stable API changes.

Now that the impl is generalized it might be more proper to live in a different file, but I did this from the web editor.

[rustbot]

r? @Noratrieb

rustbot has assigned @Noratrieb.
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

[Noratrieb]

r? libs-api

[lolbinarycat]

I actually ran into this today when adding blanket impls in one of by projects (i have a subtrait of Write and i tried adding a blanket impl on Arc)

[CAD97]

On second thought — this is a bad/dangerous implementation due to impls such as Read for &[u8], which adjusts the input reference. While it's easy to accidentally think &T: Trait means T: Trait but replacing &mut self with &self, more implementations are possible than that.

It does feel like there should be a way for such forwarding to be done, but this isn't it.

[lolbinarycat]

yep, &mut &T is mutable, &&mut T is not