Template backend for ISPC

Analogous task to TVB-2892 this task results in creation of an efficient CPU backend based on the ISPC compiler, which is similar to C and CUDA, so the CUDA backend code can be adapted directly, except for RNG which should use Random123 (e.g. https://github.com/maedoc/frand123)

If time allows, load balancing with the CUDA backend within the same node would be ideal.