Fix how edge contraction handles weights

[rmcar17]

With the new refactoring for PCG weighting options, edge contraction still only handles the case for when presence of a proper cluster in a tree adds that tree weight (rather than recalculating a new weight).

[rmcar17]

Two options here. Consider we contract an edge (a,b). How do we handle conflicting weights between edges (a,c) and (b,c). What MCS would do was go through the trees and count if if either a c or b c was a proper cluster, which works but is slightly harder when dealing with things like branch lengths (maximum of ac's or bc's LCA?). What is more intuitive to me at least is maximal support (just max of the incoming branch lengths).

[rmcar17]

The latter might reduce biases from (a,b) becoming a super node.

[rmcar17]

Fixed with max function,