parallel and GPU local interactions

GPUs has local memory for each workgroup. We should split whole space into boxes and copy all atoms which sphere-of-influence overlaps particular box int that box.

make new set of template functions (not class) for solving spatial grid. ... I put it there