Cross-scenario cuts yield obviously incorrect results on 3-scenario UC

[jeanpaulwatson]

Toward 100 iterations or so, cross scenario cuts being injected into the sub-problems yields negative gaps and other very odd/wrong behaviors.

[bknueven]

Between this and #155 it seems like the LShaped code needs to be re-visited, and it should probably be better tested given the gymnastics it does to the input scenario models.