Prior calculation performance for 1000+ mutations

[dlivitz]

Had some issues with this for large samples, but since we first wrote the software there's been a new approximation for logstirling coefficients developed:

Swaine L Chen, Nico M Temme, A Faster and More Accurate Algorithm for Calculating Population Genetics Statistics Requiring Sums of Stirling Numbers of the First Kind, G3 Genes|Genomes|Genetics, Volume 10, Issue 11, 1 November 2020, Pages 3959–3967, https://doi.org/10.1534/g3.120.401575

I have adapted this to the prior calculation, and fixed the docker build file to test.

For ~1000 mutations, takes the prior calculation from nearly a minute to almost instant, with minor error:

Old: {'a': 1.2355127299648985, 'b': 3.285775232976865, 'KL_divergence': 1.0388836127098066e-05}
New: {'a': 1.2360983922422462, 'b': 3.2871978856161417, 'KL_divergence': 1.0392287138662582e-05}

For 10,000 mutations, the old calculation took 20 minutes, and now it takes seconds.

Old: {'a': 1.2716390989170585,  'b': 4.230418159369041, 'KL_divergence': 1.3294329165569607e-05}
New: {'a': 1.2718892362810572, 'b': 4.230767762902994, 'KL_divergence': 1.329728110259154e-05}

There is no difference in clustering results between the versions.