Cuda support for fancy iterators

Changes that are needed for transposing vtkm::Vec<>. These can be merged in while we resolve performance issues with transposed arrays.