Use cutoff instead of '!= 0' when graphing duals

[staadecker]

What's happening below is the middle plot comes from a run that was slightly sub-optimal (by only 1E-7). This means the duals were very very close to zero but not exactly zero. Therefore, on the left and right plots the zero duals were filtered out but in the middle they weren't which means the plots look very different. We should change the code that is filtering out the zero duals to use a cutoff (e.g. filter out all duals smaller than 1e-5) instead of filtering out only values exactly equal to zero.