Tidy-up CUDA implementation of IProductWRTDerivBase and add CUDA kernels with additional parallelism

This MR tidy-up the previous implementation of the CUDA version of the IProductWRTDerivBase operator and introduces new CUDA kernels with additional parallelism across quadrature points.

Operators/IProductWRTDerivBase/IProductWRTDerivBaseCUDA.cu

@@ -2,9 +2,12 @@
 
 namespace Nektar::Operators::detail
 {
+
+// Add different IProductWRTDerivBase implementations to the factory.
 template <>
 std::string OperatorIProductWRTDerivBaseImpl<double, ImplCUDA>::className =
     GetOperatorFactory<double>().RegisterCreatorFunction(
         "IProductWRTDerivBaseCUDA",
         OperatorIProductWRTDerivBaseImpl<double, ImplCUDA>::instantiate, "...");
-}
+
+} // namespace Nektar::Operators::detail