Make pinned/page-locked memory optional

This would slow the proj/backproj, but will allow people to use swap memory.

This can be done "easily" at Compile time.
We can have a user flag (task 1: allow user flags easily into Compile.m) that is then passed to the CUDA code similar to IS_FOR_MATLAB_TIGRE, and is used in the following if condition (each file has one):

TIGRE/Common/CUDA/Siddon_projection.cu

Lines 345 to 348 in c55a6bb

    
           if (isHostRegisterSupported & (splits>1 |deviceCount>1)){ 
        
               cudaHostRegister(img, (size_t)geo.nVoxelX*(size_t)geo.nVoxelY*(size_t)geo.nVoxelZ*(size_t)sizeof(float),cudaHostRegisterPortable); 
        
           } 
        
           cudaCheckErrors("Error pinning memory");

The flag could be ENABLE_PINNED_MEMORY which can be set to TRUE by default.

This also needs to be used in the part where the memory is "unpinned", in the end of the main function.

	if (isHostRegisterSupported & (splits>1 \|deviceCount>1)){
	cudaHostRegister(img, (size_t)geo.nVoxelX(size_t)geo.nVoxelY(size_t)geo.nVoxelZ*(size_t)sizeof(float),cudaHostRegisterPortable);
	}
	cudaCheckErrors("Error pinning memory");