Node repeats for ReCom should reselect the pairing

[drdeford]

Instead of picking a new root for the tree, which is irrelevant for the current formulation, if none of the edges in the tree are cuttable the method should instead select two new districts to merge. One option would be to use node_repeats for the number of new trees to draw for the current pairing before selecting a new pair of districts to merge.

[drdeford]

I've tested this out a bit and somewhere around 20 seems to work pretty well.

edit: in terms of minimizing false positives