Map mutations using Sankoff's algorithm

[hyanwong]

Not a priority, but a nice comp-sci student project. At the moment, the map_mutations code uses the Hartigan algorithm, in which all mutations are equally likely. The Sankoff algorithm is an equivalent one that allows mutations to be weighted, which would be useful when looking at e.g. transitions vs transversions, or mutations in viral sequences. @jeromekelleher cooked up an example implementation in #99 and it's also mentioned in

tskit/c/tskit/trees.c

Line 7214 in d1d5bb7

* use a general cost matrix, in which case we'll use the Sankoff algorithm. For

, but I can't find any GH issue, so I thought I should open one.

There's a good description in http://profs.scienze.univr.it/liptak/ALBioinfo/2013_2014/files/PhylogenetikSkript2009.pdf (sect 4.4, which comes immediately after a description of the Hartigan algorithm). We probably want to keep the old code too, as it has better scaling properties, I think. However, we could optional pass a mutation matrix in to map_mutations, and if it is present, switch to Sankoff.