xmss: use numba JIT for poseidon2 permute

[unnawut]

🗒️ Description

This PR optimizes keygen for 12 keys from ~8 hours down to ~1.5 hours using Numba JIT, basically compiling specific python functions to machine code. The conversion does not support all operations so some changes are needed:

• Move _external_linear_layer() and _internal_linear_layer_jit() out of class Poseidon2Params so they can be compiled by Numba (with @njit annotation). The implementations are only slightly changed.
• Move permute() implementation out of class Poseidon2Params, also so that it can be compiled by Numba. Function callings changed a bit but the logic stays the same.
• The biggest change is instead of @ _M4_T directly which Numba doesn't support, we need to add _m4_multiply() to do the matmul ourselves.

A single keygen now takes about 35 minutes. There are a few more optimizations to get a single keygen down to 15-20 minutes but the code diverges more from its previous form so I opted them out. Listed down here in case we want to consider them in later:

1. Replace NumPy array ops with loops: E.g. turning state[:] = ... into for loops. I think this one would half the keygen time but the code will look different.
2. Inline the M4 matrix math: Replace the matrix and matmul with a directly derived function. But we then we lose the matrix visibility into this:

   for c in range(chunks.shape[0]):
       a, b, cv, d = chunks[c, 0], chunks[c, 1], chunks[c, 2], chunks[c, 3]
       s = (a + b + cv + d) % p
       result[c, 0] = (s + a + 2 * b) % p
       result[c, 1] = (s + b + 2 * cv) % p
       result[c, 2] = (s + cv + 2 * d) % p
       result[c, 3] = (s + 2 * a + d) % p

3. Parallelize epochs within a single keygen: Right now we parallelize at the key level so the time is always ceil(num_keys / num_cores) * single_keygen_time. We could switch to parallelize at epoch level to get single_keygen_time / num_cores * num_keys but I'm not sure if the epoch compute can maximize the cores all the time. Plus the change will be all over multiple files so maybe I can try this out after this PR.

The test in test_permutation.py should still cover this so no changes added/updated.

🔗 Related Issues or PRs

✅ Checklist

• Ran tox checks to avoid unnecessary CI fails:

  uvx tox

• Considered adding appropriate tests for the changes.
• Considered updating the online docs in the ./docs/ directory.