Goblin Translator concatenated polynomials parallel construction optimization

[Rumata888]

Goblin Translator construction of concatenated polynomials right now uses at most 4 cores (1 concatenated polynomial per core). We can use more if we want to by splitting the polynomials themselves and copying chunks in parallel instead.

Extra range constraint numerator suffers from the same issue

[Rumata888]

The construction of concatenated polynomials now uses more cores.

[ledwards2225]

@Rumata888 so this should be closed then?

[Rumata888]

Concatenated have been sped up, but ordered still use 4 cores

[suyash67]

Making ordered polynomial computation use more than 4 cores doesn't lead to any significant improvement, I tried it in this PR. The copy phase is fast enough that the parallelism overhead isn't worth it at current circuit sizes. Hence, closing this issue without merging the PR. If ordered polynomial construction becomes a bottleneck at larger sizes, we can revisit this.