RF: Change signature of `calculate_sub_jte`/`calculate_sub_grad` functions to remove CUDA coupling
Change the signature of the CostFxnSplineUtils::calculate_sub_jte
and calculate_sub_grad
functions to return their results as a std::vector
, rather than being passed a float*
to device memory. This change allows the CostFxnSSDBiasFieldMaskedExcluded
and CostFxnSSDWarpFieldSymmetricMaskedExcludedUtils
modules to be compiled with g++
rather than nvcc
.