Update Pedersen::hash_buffer once we have implemented Poseidon

[zac-williamson]

The current pedersen::hash_buffer method, which hashes a buffer of bytes is non-canonical. It splits the bytes into 31-byte field elements and then performs a pairwise hash.

i.e. we compute H(elements[0], H(elements[1], ..., H(elements[n-1]))) instead of H(elements[0], elements[1], ..., elements[n-1])

We do this in order to keep the UltraPlonk recursive verificaion circuit under 2^19 constraints (pairwise Pedersen hashes are very efficient due to using Plookup tables. Larger hashes cannot as we cannot spare the table space for more than 2 generators).

Once Poseidon is implemented, it should be more efficient than Pedersen when hashing large amounts of data. We can therefore replace current uses with Poseidon, and make pedersen::hash_buffer's behavior more canonical, or potentially remove it altogether.

[kevaundray]

Related to this comment in the pedersen cleanup issue -- Update: This is roughly still the case after the pedersen refactor