CUDA 8.0 compatibility of cuda_fixes branch

[Former user]

The header finite-volume/source/cuda/misc.h defines an atomicAdd (double*,double), which is in conflict with the built-in atomic add function available in CUDA 8.0 (https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#changes-from-previous-version).

[Jed Brown]

If you're interested in the CUDA version, you should follow @nsakharnykh's repository: https://bitbucket.org/nsakharnykh/hpgmg-cuda (which is actively developed and fixed this issue months ago). @swwilliams and @nsakharnykh, should we get rid of the cuda_fixes branch in hpgmg/hpgmg or update it?