Resolve Miri's concerns around unsafe code

[nathanwhit]

Fixes #189.

This PR essentially just ports the changes from anyhow (initial PR, cleanup commit, follow-up PR, follow-up PR) in one go.

The general theme of the changes is to use raw pointers for as much of the vtable shenanigans as possible to avoid the assumptions that references carry.

[zkat]

Gosh. This is a big PR. Thank you!

It's unfortunate that I mostly don't understand this so I just hope all this extra unsafe is... safe enough? I don't really know how to review this otherwise, because it's mostly over my head. I'm really looking forward to the Provider API so I don't have to have this Anyhow clone sitting around inline :\

Is it easy to add miri scanning to github issues? We may as well add it to CI now that you've gone through all the effort to make sure it actually passes there?

[nathanwhit]

It's unfortunate that I mostly don't understand this so I just hope all this extra unsafe is... safe enough? I don't really know how to review this otherwise, because it's mostly over my head.

Understandable, unsafe is really tricky! That's why I tried to keep the changes here as close to the anyhow fix as possible (I trust dtolnay more than I trust myself 😅).

I'm really looking forward to the Provider API so I don't have to have this Anyhow clone sitting around inline :\

+1! I'm hoping stabilization is not too far out, it'll make things a lot cleaner.

Is it easy to add miri scanning to github issues? We may as well add it to CI now that you've gone through all the effort to make sure it actually passes there?

Yup, that makes a lot of sense. I've pushed a commit to run miri in CI (you can see a successful run in my fork ) and fixed the linter warnings. For reference, before this PR around half of all of the tests were failing due to UB being detected; after this PR, they're all passing! 🎉

[zkat]

alright, let's roll with this and see what happens!