Skip set_selected_impl() of post_optimize_weight when target generic …

…layer is already created (openvinotoolkit#15852)

src/plugins/intel_gpu/src/graph/graph_optimizer/post_optimize_weights.cpp

@@ -54,7 +54,7 @@ void post_optimize_weights::optimize_weights(T& node, program& p) {
 
             // Don't run impl selection to avoid double compilation of reorder kernels
             // in main program and internal program for constant propagation
-            if (!g_node.is_constant()) {
+            if ((!g_node.is_constant()) && (!reorder.second)) {
                 g_node.set_selected_impl(g_node.type()->choose_impl(g_node));
                 if (auto impl = g_node.get_selected_impl()) {
                     auto kernel_ids = p.get_kernels_cache().add_kernels_source(impl->get_kernels_source());