LE2 variance in computation length

[siwelwerd]

I don't think this is easy to fix, but the computational length varies widely on LE2 problems. Some of the ones with 5 rows are very lengthy, while I just stumbled across one that required a single row operation to get into RREF.

[StevenClontz]

Similar to #22 (comment) I think the solution is to take an RREF matrix and undo it by rolling six row operations that don't cancel with themselves.