Support windows in stats API that do not span whole genome

[jeromekelleher]

Currently the stats API requires that we cover the entire genome with windows, which is restrictive. In particular, it prevents us from parallelising along the genome in a simple way.

The main roadblock for this is the low-level implementation of the stats, which use incremental algorithms that don't support starting and stopping easily.

This would be an excellent application for the approach discussed in #2778

[jeromekelleher]

Note this will have to take into account the issue discussed in #2792, where we need to think about the consequences of edges not being inserted into the tree in time-sorted order when we seek to a given position. Almost certainly the right thing to do for the stats API will be to do a postorder traversal (but this requires that we keep the full quintuply linked tree)

[benjeffery]

Closing for inactivity and labelling "future", please re-open if you plan to work on this.