Support go-rapidsnark in addition to snarkjs with a flag (Part 1)

[syntrust]

• Add --miner.zk-prover-impl flag to support overriding ZK prover implementation, 1: snarkjs, 2: go-rapidsnark, default to 1 （discussed on the Devs Meeting EthStorage Devs Meeting #66 Agenda pm#71)

[syntrust]

Some initial test results (time measured in seconds) :

MBP 2.9 GHz 6-Core Intel Core i9

Implementation	Proof per sample (mode 1)	Proof for 2 samples (mode 2)
snarkjs	43.8	60.1
rapidsnark + wazero.Circom2WZWitnessCalculator	47.1	82.3
rapidsnark + wasmer.Circom2WitnessCalculator	49.5	101.8

AX101

Implementation	Proof per sample (mode 1)	Proof for 2 samples (mode 2)
snarkjs	12.6	24.4
rapidsnark + wazero.Circom2WZWitnessCalculator	5.7	14.1
rapidsnark + wasmer.Circom2WitnessCalculator	10.2	17.5

Note rapidsnark supports 2 witness calculator implementations which wazero wins.

Will continue research to see if there is any optimization to do with Mac.

[syntrust]

With -tags rapidsnark_asm AX101 has an obvious improvement, esp. mode2：

Implementation	Proof per sample (mode 1)	Proof for 2 samples (mode 2)
rapidsnark + wazero.Circom2WZWitnessCalculator	5.7	14.1
rapidsnark + wazero.Circom2WZWitnessCalculator + -tags rapidsnark_asm	4.6	8.2

But this is not the case for Mac. MacOS build has the optimization enabled according to the go-rapidsnark doc.

[qzhodl]

Some initial test results (time measured in seconds) :

MBP 2.9 GHz 6-Core Intel Core i9

Implementation Proof per sample (mode 1) Proof for 2 samples (mode 2)
snarkjs 43.8 60.1
rapidsnark + wazero.Circom2WZWitnessCalculator 47.1 82.3
rapidsnark + wasmer.Circom2WitnessCalculator 49.5 101.8
AX101

Implementation Proof per sample (mode 1) Proof for 2 samples (mode 2)
snarkjs 12.6 24.4
rapidsnark + wazero.Circom2WZWitnessCalculator 5.7 14.1
rapidsnark + wasmer.Circom2WitnessCalculator 10.2 17.5
Note rapidsnark supports 2 witness calculator implementations which wazero wins.

Will continue research to see if there is any optimization to do with Mac.

By running proof by using rapidsnark(https://github.com/iden3/rapidsnark) vs snarkjs, I can reproduce the performance gain (8s vs 40s)