Benchmark witness generation separately

[brech1]

Description

Related to overhead in current proving benchmarking, for example:

pub fn run_prove(&self, input_exp: u32) -> NoirProof {
    let proof_scheme = self.proof_schemes.get(&input_exp).unwrap();
    let toml_path = self.toml_paths.get(&input_exp).unwrap();
    let witness_map = proof_scheme
        .read_witness(toml_path.to_str().unwrap())
        .expect("Failed to read witness");
...

We should remove any execution not related to the proving in that scope and include a separate witness generation benchmarking for all benchmarks.