Fix reciprocal polarization energy with PME

[andysim]

The reciprocal polarization energy was missing the Coulomb constant, which is fixed by this patch. I added the factor as electric / dielec, but I noticed that many places, like this, only electric is used.

I also had to roll back this change or the code gets stuck in an infinite loop on my laptop (macOS 10.12, gfortran 7.1).

[jayponder]

This is now fixed. The code is somewhat different for the "energy only" vs. "energy & gradient" routines. In the "energy only" versions the "f" factor (ie, electric/dielec) is applied to the final energy value after the accumulation loop. In the "energy & gradient" versions, "f" is applied to the "fphi" potential and field values, and then propagates through the full energy and gradient loops.

[andysim]

Great, thanks Jay. Given the different code paths, perhaps I should add extra analyze tests with 'E' and 'EV' to test both energy and gradient routines.