Fr FFT doesn't really use all the cores fully

[nazar-pc]

I was not able to find the culprit, but on my machine (24C/32T 13900K) switching from single-threaded to parallel FFT uses multiple CPU cores, but only slightly, resulting in ~3x time reduction, likely with a lot of wasted compute in the process.

It would have been nice to leverage CPU cores fully. There is already short-circuiting for parallelism when number of elements on each side is below 256:
https://github.com/sifraitech/rust-kzg/blob/5655cdd039788b1df4d628036fdc705080e000eb/blst-from-scratch/src/fft_fr.rs#L40

Tweaking it either way worsened time for me though.

I think there must be an opportunity for further performance or at very least efficiency improvements.