Add parsimony_score function

[jeromekelleher]

We would like a function to compute the parsimony score for each site in a dataset.

The implementation can be based on the work in the tskit-paper repository where there is a number of different versions implemented using numba.

We would return a new dataset which includes the variable parsimony_score, which has a value for each site in the variant data.

Initially we can use a single site version, but I think we would want to use the vectorised version for better efficiency:

https://github.com/tskit-dev/tskit-paper/blob/main/tree_performance/benchmark.py#L147

Ultimately, we'd like to use Dask to run this vectorised version on chunks of sites, in parallel (although I'm not entirely clear how this would be structured)