[Performance Improvement] Poseidon with AVX2 (x86-64)

[dloghin]

This is a performance improvement PR targeting x86-64 systems with AVX2 support.

1. target bottleneck
   Currently, Plonky2 does not have a working AVX2 Poseidon hashing implementation. AVX2 has the potential to improve the performance of Poseidon hashing by executing up to 4 64-bit Goldilocks operations at the same time.

2. how performance is measured
   We use current benchmarks in plonky2/benches:

cargo bench --bench=hashing
cargo bench --bench=merkle

With AVX2 support, we run:

RUSTFLAGS="-C target-feature=+avx2" cargo bench --bench=hashing
RUSTFLAGS="-C target-feature=+avx2" cargo bench --bench=merkle

3. characteristics of the machine used (CPU, OS, #threads if appropriate)
   We checked the performance on a c7i.2xlarge AWS instance with AMD CPU with the following specs:

CPU: AMD EPYC 9R14 (16 vCPU)
RAM: 16 GB
OS: Ubuntu 24.04
Compilers: GCC 13
Rust: rustc 1.82.0-nightly

4. performance before and after the PR
   Performance improvement is 10-12%. For example, the output of RUSTFLAGS="-C target-feature=+avx2" cargo bench --bench=merkle is:

merkle-tree<GoldilocksField, PoseidonHash>/8192
                        time:   [25.929 ms 25.984 ms 26.017 ms]
                        change: [-11.076% -10.767% -10.460%] (p = 0.00 < 0.05)
                        Performance has improved.
merkle-tree<GoldilocksField, PoseidonHash>/16384
                        time:   [53.447 ms 53.596 ms 53.724 ms]
                        change: [-10.998% -10.679% -10.362%] (p = 0.00 < 0.05)
                        Performance has improved.
merkle-tree<GoldilocksField, PoseidonHash>/32768
                        time:   [107.41 ms 107.64 ms 107.83 ms]
                        change: [-10.545% -10.229% -9.8422%] (p = 0.00 < 0.05)
                        Performance has improved.

[dloghin]

Hi Polygon team, do you have any updates on this PR? Thanks!