Remove ?Sized bounds or add CI test that they work

[ids1024]

#132 added trait bonds with ?Sized.

Context<dyn HasDisplayHandle> was mentioned there, but if you try to specify that type anywhere, it errors because Context and ContextDispatch do not have D: ?Sized bounds. So the public API doesn't actually allow that.

Adding those bounds is easy, but is there a way to construct a Context<dyn HasDisplayHandle>?

If ?Sized is useful, we should make sure it works and have a test that it continues working on all platforms. Otherwise, it can be removed, which may make other refactoring easier.

CC @notgull

[notgull]

Is there any explicit refactoring that doesn't work with ?Sized bounds?

I think it only works on Nightly, but there I think you can coerce an &Context<Window> to an &Context<dyn HasDisplayHandle>.

[ids1024]

I don't think it should be an issue with anything, but ?Sized may need to be added in more places, and some care may be needed not to break this use case, so if it's possible it would be good to have a test for it.

It's also possible to use Context<Box<dyn HasDisplayHandle>>, so the advantage is saving an extra boxing, I guess.

[ids1024]

I don't think Rust could allow a coercion from Context<T> to Context<dyn HasDisplayHandle<T>> since type type signature doesn't express exactly how the T is used.

I think if it's possible we'd at least need to provide some kind of conversion for that to work. And I don't know exactly how that works when we have a boxed/ref-counted custom unsized type containing the T.