Remove all possible panics from the main resolver

[blake-mealey]

This project does a good job of handling error cases, but there are one or two places in the main resolver code which makes assumptions about the graph which should be true but occasionally aren't (I've so far only experienced this as a result of some bugs like #104). When it panics part-way through a resolve, it can sometimes cause some major issues for the next attempt (which can wreck the graph) so these should really be handled properly (even though in a perfect world the assumptions should be true)

[blake-mealey]

Adding to milestone v1.0.0 because this could cause costly mistakes in production