[OpenMP] Fix -foffload-lto for offloading to Nvidia

Summary:
A previous patch ensured that we can't use `-foffload-lto` with CUDA in
non-RDC mode. This did not correctly check for if we were actually
offloading to CUDA so it ended up preventing us from using LTO for
OpenMP.

Author: jhuber6

clang/lib/Driver/ToolChains/Clang.cpp

@@ -4851,7 +4851,8 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                                options::OPT_foffload_lto_EQ)
                    ->getAsString(Args)
             << Triple.getTriple();
-      } else if (Triple.isNVPTX() && !IsRDCMode) {
+      } else if (Triple.isNVPTX() && !IsRDCMode &&
+                 JA.isDeviceOffloading(Action::OFK_Cuda)) {
         D.Diag(diag::err_drv_unsupported_opt_for_language_mode)
             << Args.getLastArg(options::OPT_foffload_lto,
                                options::OPT_foffload_lto_EQ)