Parallelizin using SIMD

[Fleek has done some interesting work that leverages SIMD](https://blog.fleek.network/post/fleek-network-blake3-case-study/) in blake3 hash implementation.

Given that merkle-references use pretty similar tree algorithm we may be able to leverage same technique to paralelize deriviation.

Start by reading the linked Fleek SIMD case study, then locate the merkle-reference derivation entry points in the repository. Determine whether the tree algorithm can be parallelized in this implementation and define the correctness and performance checks needed for the work to be considered done.

Written by the indexing model from the issue text.