Perf improvements for keccak,state circuit

[pinkiebell]

Description

• Improves the keccak performance for padding rows significantly
• small performance gain in state circuit with high padding rows

[CPerezz]

That makes complete sense! Indeed, I think we should have a logging feature directly probably. So that we avoid situations like this one..

WDYT @ed255 ??

Thanks for this @pinkiebell good catch!

[pinkiebell]

The greatest perf improvement came from the precomputed padding_rows. That alone improved perf about 6.7 times.
After some more digging I also see that the closure for assign_region is invoked twice with the simple floor planner. I might sense to keep that in mind to move heavy computation outside that closure and just reference it when doing the assignments.

[CPerezz]

@pinkiebell maybe makes sense to file an issue in Halo2 for that? So that we can check it deeper?