Avoid setting additional linker flags at global scope

 * Setting CMAKE_EXE_LINKER_FLAGS was used also for NMODL
   when NMODL is built from submodule
 * In case of LLVM OpenMP offload, if binary is being created
   from the object that is not compiled with OpenMP flags,
   it causes link error:

   echo "int main() { return 0; } " > foo.cpp
   clang++ -c foo.cpp                                           # adding openmp flags here doesn't cause below error
   clang++ foo.o -o foo -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda

     nvlink fatal   : Could not open input file '/tmp/foo-0b9a1a.cubin'
     clang-12: error: nvlink command failed with exit code 1 (use -v to see invocation)

CMake/OpenAccHelper.cmake

@@ -98,7 +98,6 @@ if(CORENRN_ENABLE_GPU)
   endif()
   # avoid PGI adding standard compliant "-A" flags
   # set(CMAKE_CXX14_STANDARD_COMPILE_OPTION --c++14)
-  string(APPEND CMAKE_EXE_LINKER_FLAGS " ${NVHPC_ACC_COMP_FLAGS}")
   # Use `-Mautoinline` option to compile .cpp files generated from .mod files only. This is
   # especially needed when we compile with -O0 or -O1 optimisation level where we get link errors.
   # Use of `-Mautoinline` ensure that the necessary functions like `net_receive_kernel` are inlined

CMakeLists.txt

@@ -419,6 +419,9 @@ endif()
 # =============================================================================
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${IGNORE_UNKNOWN_PRAGMA_FLAGS}")
 
+# Set coreneuron specific linker flag after NMODL submodule is added
+string(APPEND CMAKE_EXE_LINKER_FLAGS " ${NVHPC_ACC_COMP_FLAGS}")
+
 # =============================================================================
 # Add main directories
 # =============================================================================