Generic Prover

[gabriel-barrett]

This PR changes the prover API quite a bit

1. Prover is generic over a FrameLike type, no more hardcoded to the Multiframe
2. Prover is no more tied to coprocessors
3. Prover no longer has a function over Frames

This PR was motivated by the work on systems. I think the API today is too rigid and I'm trying to make it more flexible. I've removed things which I believe has no business being inside the prover, like coprocessors, since this is a property of the multiframe, which is subject to change. Frames are also not important to the prover, only the multiframe is. Plus, with coroutines we might not build a vector of frames anymore, but a more complicated structure. We will need to translate that structure into a ordered multiframe structure however. As of now, the prove function requires a Vec but we might generalize that to iterators or something

[huitseeker]

I like this in that it's a meaningful simplification of the proving traits.

We can see that there is a bit of code duplication in evaluate_and_prove and prove_from_frames between the Nova and Supernova instances of Prover. This is perhaps a sign that those two methods are meant for a common trait between each of those two structs, one that would give access to the right "frame factory" in each instance. It would be good to figure out if there isn't a way to write that trait a consider if it helps. But I don't think that needs to block the present PR either.