Roe eigensystem has a bug

Pattern of passing/failing tests suggests:

• issue is certainly related to direction (x,y,z) in which eigensystem is being considered
• issue is not present when velocity is the same in x,y,z
• issue is exposed by Roe property test, not Roe identity check, implying eigenvalues are at issue, but that's pretty simple code that seems correct
• possibly eigenvalues are out of order (cursory check suggests this is not the issue)
• could be that the way the velocities are setup is wrong (although again, a cursory check suggests this is not the issue)