chore: switch benches to hzkg/bn254

[huitseeker]

Fixes #278

[huitseeker]

!gpu-benchmark

[huitseeker]

Note to self: investigating #281, and keeping the pasta benches at hand for that reason: it's easier if I have direct access to the same benches as those that showed the difference in speed betw. GPU/GPU-less.

[samuelburnham]

!benchmark --features cuda,asm

[huitseeker]

Now that I have the data for a branch that does not have any calls to the MSM in the IPA (which shows no change), I'm comfortable merging this.

[github-actions]

Benchmark for 4cc09c8

Click to view benchmark

Test	Base	PR	%
CompressedSNARK-NIVC-Commitments-2/Prove/StepCircuitSize-514444	58.2±0.80s	27.4±0.76s	-52.92%
CompressedSNARK-NIVC-Commitments-2/Prove/StepCircuitSize-6540	14.5±0.13s	11.1±0.17s	-23.45%
CompressedSNARK-NIVC-Commitments-2/Verify/StepCircuitSize-514444	423.8±1.55ms	46.7±0.53ms	-88.98%
CompressedSNARK-NIVC-Commitments-2/Verify/StepCircuitSize-6540	83.4±0.67ms	46.5±0.43ms	-44.24%
RecursiveSNARK-NIVC-2/Prove/StepCircuitSize-514444	211.9±0.91ms	198.1±1.41ms	-6.51%
RecursiveSNARK-NIVC-2/Prove/StepCircuitSize-6540	53.9±0.36ms	45.3±0.37ms	-15.96%
RecursiveSNARK-NIVC-2/Verify/StepCircuitSize-514444	154.7±3.46ms	149.6±2.78ms	-3.30%
RecursiveSNARK-NIVC-2/Verify/StepCircuitSize-6540	34.8±0.67ms	35.6±0.64ms	+2.30%