[FRI] Dynamic layer folding

[irakliyk]

Right now, we specify a single folding factor for all FRI layers. Out of supported folding factors (4, 8, 16), it seems like 8 results in the smallest proof in most situations. However, it might be possible to optimize this even further by using different folding factors across different layers. I wouldn't expect the impact to be huge - but getting something around 5% proof size reduction may be possible.

One way to go about this is to deterministically create a schedule of folding factors based on extended domain size, and field element size, and hash function digest size. This schedule can then be provided as an input to FRI prover/verifier.