Update integrator used in OpenMM driver

[mattwthompson]

          A crucial detail I've run into while getting this ready; [OpenMM's VerletIntegrator](http://docs.openmm.org/7.2.0/api-python/generated/simtk.openmm.openmm.VerletIntegrator.html#simtk.openmm.openmm.VerletIntegrator) (and in general, any leapfrog algorithm integrator in OpenMM) **will report non-zero kinetic energies when evaluating State, _even_ if all velocities are zero**. Taken [straight from the horse's mouth](http://docs.openmm.org/7.2.0/api-python/generated/simtk.openmm.openmm.State.html#simtk.openmm.openmm.State.getKineticEnergy):``

"Note that [KE] may be different from simply mv/2 summed over all particles. For example, a leapfrog integrator will store velocities offset by half a step, so they must be adjusted before computing the kinetic energy."
``

Some quick testing shows that for static structures exported from Interchange to OpenMM, VerletIntegrator and LangevinIntegrator both yield nonzero KEs, while LangevinMiddleIntegrator and BrownianIntegrator DO yield the expected 0 KE. I haven't yet checked the full list of supported Integrators exhaustively.

This becomes relevant to Interchange when noting that the drivers.openmm layer does use VerletIntegrator in its evaluation, and does not check 0 KE as far as I can tell. This has been introducing errors into my current data, and is likely doing the same for yours @mattwthompson

Originally posted by @timbernat in #1086 (comment)