feat(nifs): impl `protogalaxy::poly::compute_G`

[cyphersnake]

Motivation Close #260

Overview

• Implementation of compute_G via tree_reduce
• Multithreaded implementation of fold_witness
• Single-threaded fold_plonk_challenges implementation (there are not so many items, so there is no sense to use rayon)
• An implementation of multi_product, analogous to itertools, but without Clone.
• An implementation of try_multi_product, analogous to itertools, but with processing Result in Item. This allows not to finish processing when the first error occurs.
• Allocate the folded_trace module to control the size of protogalaxy::poly

[cyphersnake]

The algorithm separated from the context is here, for your reference

https://gist.github.com/rust-play/89ae0609d30feb22d2585a6f43ccc07e

[cyphersnake]

I did not take into account that PlonkWitness is not a matrix, in general it is not hard to correct by making the indexes trickier

I will postpone this task for now due to higher priority, however, the fix will be not very large

[cyphersnake]

Benchmark data is still being collected, but there doesn't seem to be any code left there, I'm going back to this task

[cyphersnake]

Generally fixed, but because of cloning there is multiple rows traversal, which can be annoying. It is also possible to create result vectors more optimally.

But in order not to delay, I suggest to return to this optimization, if it becomes bottle neck

[cyphersnake]

@chaosma If you don't understand this big iterator, let me know, we may need a call to write a good doc to it together. It's really not very intuitive, but I had to sacrifice understandability for optimality

[chaosma]

@chaosma If you don't understand this big iterator, let me know, we may need a call to write a good doc to it together. It's really not very intuitive, but I had to sacrifice understandability for optimality

I think I got it. Other comments are addressed. I leave a new comment there.