[C-API] Bugfix for ml_pipeline_element_get_handle

Need to ref element obtained by iterate_element() and unref it in cleanup_node() for common_elem

- common_elem by ml_pipeline_element_get_handle() need to unref in cleaunup_node() because g_hash_table_lookup assign new memory.
  but normal operation, unref is not necessary, so add ref in iterate_element()
- Now, the element's finalize function is called after ml_pipeline_element_get_handle()

Signed-off-by: hyunil park <hyunil46.park@samsung.com>

c/src/ml-api-inference-pipeline.c

@@ -601,6 +601,8 @@ cleanup_node (gpointer data)
   if (e->sink)
     gst_object_unref (e->sink);
 
+  g_object_unref (e->element);
+
   _ml_tensors_info_free (&e->tensors_info);
 
   g_mutex_unlock (&e->lock);
@@ -915,7 +917,8 @@ iterate_element (ml_pipeline * pipe_h, GstElement * pipeline,
               if (element_type != ML_PIPELINE_ELEMENT_UNKNOWN) {
                 ml_pipeline_element *e;
 
-                e = construct_element (elem, pipe_h, name, element_type);
+                e = construct_element (gst_object_ref (elem), pipe_h, name,
+                    element_type);
                 if (e != NULL) {
                   if (g_str_equal (element_name, "tensor_if"))
                     process_tensor_if_option (e);