Update and clean up hash benchmarks

With 0xMiden/crypto#720, we are now using Plonky3 backend for field implementation. We should update our [hash benchmarks](https://github.com/0xMiden/crypto/tree/next/miden-crypto/benches#hash-functions).

I have done this already for an M4 Max chip, and the changes were as follows:

- RPO: 2.9 µs -> 2.5 µs
- RPX: 1.5 µs -> 1.3 µs
- Poseidon2: 0.7 µs -> 0.7 µs (basically unchanged).

We should run this for other planforms and remove the platforms that are no longer relevant. I think the platforms we care about are:

- M1 - a good baseline benchmark.
- Graviton4 - something we run in AWS for remote provers (supports SVE acceleration).
- AMD Zen5 (e.g., AWS c8a instance) - we could also run these for remote provers (supports AVX512 acceleration).

It probably makes sense to re run these once 0xMiden/crypto#732 is implemented.

One open question is whether Goldilocks field in Plonky3 can take advantage of SVE and AVX512 acceleration (cc @Nashtare and @sai-deng in case you guys know).