Gas optimization through batch-invert barycentric denominators in sumcheck

[nicolaslara]

When working on https://github.com/nicolaslara/solana-noir-verifier/ I had to do a few optimizations to reduce the compute time. Batch inversions was one of them. I noticed these were missing upstream, so adding the highest impact one here.

Here are the performance changes from the tests I ran:

Test	Before	After	Saved	%
Add2	968,225	922,589	45,636	4.7%
Add2 ZK	1,885,965	1,834,205	51,760	2.7%
Blake	1,461,863	1,346,344	115,519	7.9%
Blake ZK	2,451,954	2,321,207	130,747	5.3%
ECDSA	1,528,814	1,405,416	123,398	8.1%
ECDSA ZK	2,526,897	2,387,271	139,626	5.5%
Recursive	1,742,905	1,587,720	155,185	8.9%
Recursive ZK	2,773,343	2,597,880	175,463	6.3%

The next high impact version of this would be to apply it to the gemini fold. I can try that if you find this valuable