Make box wrap reversible to original coordinates and support molecules with length > 1/2 box length

[GregorySchwing]

Is your feature request related to a problem? Please describe.
I am unable to achieve complete passing of the checkpoint unit test for certain systems, particularly the BPTI and K Channel systems. In these systems, the unwrapped trajectory coordinates are matching between a single simulation and a simulation which stops and continues from checkpoint. However, the internal wrapped coordinates are not matching when comparing the positions after running 1/2 the uninterrupted simulation and immediately after continuing the checkpointing simulation. The indication that this is related to box wrapping is the velocities do match, which are not operated on by the box wrapping functionality. The problem is due to the stateless box wrap, which simply checks if a coordinate is > 1/2 the box length or < 0. If so, it moves it into the image. It does not however, track how many images that coordinate was from the current image. Furthermore, molecules which exceed the length of 1/2 the box length will never be entirely in the same image.

Describe the solution you'd like
Implement a transformation vector for each atom which is a multiplier in each dimension <x,y,z> of box length in each direction for reversible box wrapping.

Describe alternatives you've considered
I have tried positioning the simulation box entirely in the positive, positive, positive quadrant of 3D space, with the center of the box between 0 and 1/2 the box length in each direction. This still didn't produce consistent wrapped coordinates K_Channel between a single simulation and a checkpointed simulation, which has been unwrapped and rewrapped, at the same step number.