Add return type to `Yoke::with_mut` and `Yokeable::transform_mut`

[jdygert-spok]

I believe this is still sound since R can't involve 'b.

fn with_mut<'a, F, R>(&'a mut self, f: F) -> R
where
    F: 'static + for<'b> FnOnce(&'b mut <Y as Yokeable<'a>>::Output) -> R;

fn transform_mut<F, R>(&'a mut self, f: F) -> R
where
    F: 'static + for<'b> FnOnce(&'b mut Self::Output) -> R;

This would be a breaking change, mainly for anyone manually implementing Yokeable.

Notably, this lets us easily call Iterator::next on an iter in a yoke, or even implement Iterator for Yoke.

[sffc]

My initial reaction is that I think this is a beneficial change. Returning things from closures is a pain, especially closures with restrictive lifetime parameters like the ones here. Does R need to be 'static?

CC @Manishearth

[Manishearth]

My immediate reaction is worry about contravariance and implied bounds sneaking in a 'b anyway: see rust-lang/rust#106431 for the last time this kind of thing happened. It gets pretty subtle and is hard to reason about.

I am overall in favor of this if we can be really sure that it works the way we expect it to.

[Manishearth]

R: 'static would indeed be a way to make this work without any chance of problems. Would be nice to allow self-borrows, but R: 'a is almost certainly going to get the implied bounds messiness.

[sffc]

I think R: 'static solves a lot of use cases; it allows the closure to return primitives and owned data, which it can't currently do since the closure is required to be 'static, so it can't save output variables to its context.

https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=a9d03f0a3d04f8a6e7e212486bd66858

[Manishearth]

I'm in favor of R: 'static + a breaking change

[jdygert-spok]

Hmm, that variance issue is quite tricky. My gut says it wouldn't be a problem here since F is covariant over R, but I don't really know how to verify the soundness.