feat(nifs): parallel `tree_reduce` in `protogalxy::poly`

[cyphersnake]

We use itertools::tree_reduce to optimize, as described in the protogalaxy paper, the counting of F & G

This algorithm can be paralleled , and we can do so in a generalized way

[cyphersnake]

And we can make the `try_tree_reduce' method to stop after the first error occurs