Use junction trees to find best haplotypes

[davidbenjamin]

Once we have built junction trees for linked de Bruijn graphs we can use them to find phased haplotypes that handle repeats / cycles. This essentially amounts to running the current Dijkstra's algorithm on junction trees (which are, after all, DAGs) instead of SeqGraphs. That is, the edge weights can remain log branching ratios.

The complication is that while using the the "oldest" junction tree for edge weights we must also traverse younger junction trees in order to rely on them downstream.

We may or may not want to work out a rule for when to use a younger junction tree with much higher coverage than an older one, or to combine multiple trees into a single vote.