triclinic support in long-range solvers

Hi all - released a patch today with a feature
some users have asked for, for a long time.

Stan Moore (Sandia) has done some nice work
to add support for triclinic (non-orthongal boxes)
to several of the long-range KSpace solvers.

Namely PPPM, MSM, and Ewald. Previously
only the ewald/disp solver supported this
(traditional Ewald and/or LJ dispersion).
Code for this needs to be migrated into the
derivate solvers, such as USER-CUDA, USER-OMP,
pppm/tip4p, etc. So for now, those will still flag triclinic
boxes with an error.

See the 12 May 13 patch for details …