long-range solvers with triclinic box

I think I fixed the major issue here, and the bugfix should be released soon. Basically Ewald/PPPM/MSM needed to take the absolute value of the tilt factors for the error estimator. FYI, the error estimator for ewald/disp doesn’t yet take into account triclinic skew like it should; when I get some time I’ll fix that as well.


In other words the “bad” results were actually
not due to an error in the long-range calculation
itself, but rather that with the negative tilt factor,
the setup part of the code chose a coarser (less accurate)
grid than it should have.