Use strongly typed wrapped indices in VecDeque

[Kobzol]

This is far from perfect, but it would have prevented #151769.

r? @joboet

[joboet]

I love the general approach, though I have a couple of nits. And personally I'd chose BufferIndex over WrappedIndex, but I'm fine with whichever you choose.

View changes since this review

[joboet]

I don't think this safety invariant is particularly useful, as all functions that act on the buffer still need to require that the WrappedIndex is specific to that particular VecDeque with a given capacity.

[Kobzol]

Yeah, it's a relatively weak invariant. That being said, I still thought it would be a good idea to provide a "SLOW DOWN" sign when someone wants to conjure the index out of the blue, which is why I marked the function unsafe.

[joboet]

The return type should be usize, right? It isn't a buffer index but rather a length.

[Kobzol]

Right, the type represents something like "value in-bounds". My thought process was that if we have two in-bounds values, and we do an absolute diff of them, the result also has to be in-bounds (w.r.t. the same VecDeque). But we do treat it as usize everywhere the function is currently used, so.. I'll change it to usize.

[Kobzol]

I did name if BufferIndex index originally 😆 I guess both works.

[Kobzol]

I do slightly prefer the Wrapped name, I think that the buffer part is implied by the word "index". The main idea is to express that the index was wrapped to not go over bounds, which is how it is created through most means.

@rustbot ready