Enhance mem_usage_benches to more precisely measure peak memory usage

[ounsworth]

Currently, the memory usage benchmarks are structured like this:

fn main() {
    bench_mldsa87_lowmemory_sign()
}

fn bench_mldsa87_lowmemory_sign() {
    use bouncycastle::mldsa_lowmemory::{MLDSA87, MLDSA87_SK_LEN, MLDSA87PrivateKey, MLDSATrait};

    eprintln!("MLDSA87_lowmemory/Sign");

    let sk = MLDSA87PrivateKey::from_bytes(&[
        0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E,
        0x0F, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D,
        0x1E, 0x1F,
    ])
    .unwrap();

    let msg = b"The quick brown fox jumped over the lazy dog";

    let mu = MLDSA87::compute_mu_from_sk(&sk, msg, None).unwrap();
    let sig = MLDSA87::sign_mu_deterministic(&sk, &mu, [0u8; 32]).unwrap();
    print!("{:x?}", sig);
}

Then we run this main within valgrind to measure its peak memory usage.

The problem with this approach is that we are also measuring memory usage of main(), the seed and msg constants, and the two print! macros. (the prints, I think, are required to force the compiler not to elide the actual operations) For some algorithms like sha3, these overheads It would be more ideal to measure only the stack / heap memory usage of the ::signmu_deterministic(..), but I have no idea how to do this. Needs some research.