Preliminary implementation of multi-gpu communication for nonbonded force calculation; energy kernel not yet implemented

Don't use warp aggregated intrinsics; the compiler for cuda9 and on can do a better job\n\nMaybe older versions of cuda would also perform better

Initial port to cuda11: include c++14 std, use texture objects instead of deprecated texture refs, squash warnings

Added 4-particle vecangle potential; untested, and configuration.cpp still needs to build the potential dictionary using type,filename tuples as keys

Changed BondAngle to be the product of the energy of potentials applied to the central bond and two angles formed by four particles