Internalize multi-tree comparisons in C++

[ms609]

When comparing all pairs of trees, we could attain faster results by:

• Loading all trees into C++ and converting to split lists once (rather than for each pair)
• Storing a sorted list of splits alongside a list of their properties
  • Use a k-way merge to produce a single index of all unique splits
  • Each tree will then be represented as a series of links to splits
  • Each unique split can have its properties (in_split) calculated and stored once
  • Also possible to compare all pairs of splits once -- if this doesn't consume too much memory.