CUDA: Now pass correct FLAGS when device link cuda executables.

Previously we had a two issues when building cuda executables
that required separable compilation. The first was that we didn't
propagate FLAGS causing any -arch / -gencode flags to be dropped, and
secondly generators such as ninja would use the CXX language flags
instead of CUDA when the executable was mixed language.

Source/cmLinkLineDeviceComputer.cxx

@@ -59,6 +59,12 @@ std::string cmLinkLineDeviceComputer::ComputeLinkLibraries(
   return fout.str();
 }
 
+std::string cmLinkLineDeviceComputer::GetLinkerLanguage(cmGeneratorTarget*,
+                                                        std::string const&)
+{
+  return "CUDA";
+}
+
 cmNinjaLinkLineDeviceComputer::cmNinjaLinkLineDeviceComputer(
   cmOutputConverter* outputConverter, cmStateDirectory stateDir,
   cmGlobalNinjaGenerator const* gg)