zkevm: add ECRECOVER, SHA2-256, IDENTITY and RIPEMD precompiles

[jsign]

This PR:

• Adds tests for new precompiles: ECRECOVER, SHA2-256, IDENTITY, and RIPEMD.
• Removes zkevm pytest markers since the testing framework now automatically does this.
• Uses Environment() configured gas limits, which are expected to be correctly set via the new --block-gas-limit filling flag.
• I updated the CI pipeline to use 36M gas limit for filling.

Cycle counts in the new precompiles:

• ECRECOVER: 682 million cycles
• SHA2-256: 5.8 billion cycles
• IDENTITY: 305 million cycles
• RIPEMD: 360 million cycles

The last three are all part of a generic test since these precompiles have the same input parameter structure (i.e., data). The generic test automatically discovers the optimal input length for each case to amortize the static cost as much as possible against the quadratic cost of memory expansions. ECRECOVER is a separate test since the parameter input structure is particular.

[jsign]

The gas_limit in all tests is configured via Environment().gas_limit. Since by default the testing framework sets as twice the current mainnet one, that looks unfair/unexpected for zkevm tests.

Thus, I think it's much better to define the expectation via --block-gas-limit for the filling.

[jsign]

bytes_per_unit_of_work is mainly a concept of for how many bytes of input the precompile has to do "computational work". In the case of hashes, is related to the block size of the compression function. For identity is "1" since work is linear (i.e., not "chunked")

cc @kevaundray

[jsign]

I didn't find a way to pull the static and dynamic costs from precompiles. Is there a way to do this? Mostly asking for future-proofness

[chfast]

These are not zkevm specific but good general benchmarks for EVM. Can we make this framework for benchmarks a bit more generic? I'd like to contribute my existing benchmarks and more.