Use Faster STARK Configuration for Testing

[sai-deng]

We can use a faster STARK configuration for testing, which will help speed up some longer tests, such as test_segment_proof_generation_without_keccak(), which requires many recursions.

[sai-deng]

Currently, when I reduce the num_query_rounds in the fri_config, I encounter errors related to common data mismatches, and the final table proofs are missing a CosetInterpolationGate.

[Nashtare]

It'd be nice to enable a customizable config through the CLI as well for e2e testing of the prover flow in a cheap way, and not just for the evm_arithmetization crate. Could probably be as simple as (assuming conjectured_security of FRI as plonky2 does), specifying a targeted security level and have the application create the config (the only tweak would be on the number of queries most likely, eventually the pow step if too high too).