Dollo parsimony approach sub-optimal

[jeromekelleher]

We are potentially using too many back mutations when placing non-inference sites. We should try to find clades of ancestral state samples and put mutations over those rather than just putting a back mutation over every sample.

Fix, and update docs in inference.rst.

[hyanwong]

I think this can be closed, @jeromekelleher ?

[jeromekelleher]

Closed in #179 and #180.