GitLab

So far tested only on GPU-having windows machine.

Turn default to ON for nightly building Cuda FreeEnergy plugin, since it is desired for pyopenmm release.

Accumulation of Born force was incorrect in USE_OUTPUT_BUFFER_PER_WARP block w/ cell coordinates  x != y

Still doesn't build on windows, but might help with linux (todo next)

Cast exponent in pow() function to RealOpenMM, to avoid overload ambiguity error on Windows when RealOpenMM is defined as double.

Exported some symbols from Cuda plugin, for use by external NML plugin.
  cudaOpenMMInitializeIntegration, gpuSetConstants, CudaPlatform
  (but I couldn't figure out how to export kGenerateRandoms)
Also turned off TestRandomCuda on Windows, since it always fails and makes the dashboard look bad.
Renamed InitializeIntegration method to cudaOpenMMInitializeIntegration, now that symbol is being exported.

Increased the tolerance on two tests that started failing when we switched the reference platform to double precision

Associated code will be removed once confirmed deletions do not break builds

Code surrounding 'RealOpenMMType == 0' will be removed 
once confirmed free-energy plugin code does not break

move the plugin to its own project.
This change might also simplify pyopenmm nightly test issues.