feat: Expose NewDefaultPermutation in Poseidon2 packages

[Tabaie]

For use as a compression function and not a range-extended hasher

---

Note

Expose NewDefaultPermutation across Poseidon2 packages and switch Merkle–Damgård hashers to use it; add SetBytesCanonical property tests and koalabear Poseidon2 test vectors.

• Poseidon2:
  • Add NewDefaultPermutation() returning Permutation{GetDefaultParameters()} in poseidon2.go for multiple curves/fields and update NewMerkleDamgardHasher() to use it.
  • Adjust hasher buffer sizing to derive from permutation params; update generator templates accordingly.
• Tests:
  • Add property-based test: SetBytesCanonical(Bytes()) remains constant in fp/fr element tests across curves and in field generator templates.
  • Add koalabear Poseidon2 test vectors (poseidon2/test-vectors.json) and vectors_test.go validating width, block sizes, and output.

^{Written by Cursor Bugbot for commit 413eea9. This will update automatically on new commits. Configure here.}

[YaoJGalteland]

The discrepancy seems to arise from differences in how the uint32 values are being encoded into field.Element objects and then serialized into byte buffers.

• When using this encoding:
  _, err = binary.Encode(buf[i*koalabear.Bytes:(i+1)*koalabear.Bytes], binary.BigEndian, in[i])
  I printed the input to Permutation (the place right before h.Permutation in (h *Permutation) Compress), which is:
  [[0] [0] [0] [0] [0] [0] [0] [0] [0] [33554430] [67108860] [100663290] [134217720] [167772150] [201326580] [234881010]], it is equal to [[0] [0] [0] [0] [0] [0] [0] [0] [0] 1xOne(), 2xOne(), 3xOne(),...].

• When using this encoding:

elem := koalabear.Element{in[i]}
elemBytes := elem.Bytes()
copy(buf[i*koalabear.Bytes:(i+1)*koalabear.Bytes], elemBytes[:])

the input to permutation is: [[0] [0] [0] [0] [0] [0] [0] [0] [0] [1] [2] [3] [4] [5] [6] [7]].

I think the input transformation depends on how you interpret and encode the uint32 values into the field.Element representation.

[YaoJGalteland]

_, err = binary.Encode(buf[i*koalabear.Bytes:(i+1)*koalabear.Bytes], binary.BigEndian, in[i])
and

elem := koalabear.NewElement(uint64(in[i]))
elemBytes := elem.Bytes()
copy(buf[i*koalabear.Bytes:(i+1)*koalabear.Bytes], elemBytes[:])

produce equivalent results. So the main difference from the first line is how in[i] is converted to a koalabear.Element.