Fix extract workflow from history with empty mapped over collections

Fixes #18484.

It's not ideal on the one hand, but better to have a workflow with gaps
than an internal server error IMO. We do the same for datasets in some
non-terminal states.

lib/galaxy/webapps/galaxy/controllers/workflow.py

@@ -320,6 +320,7 @@ def build_from_current_history(
             # Optionally target a different history than the current one.
             history = self.history_manager.get_owned(self.decode_id(history_id), trans.user, current_history=history)
         if not user:
+            trans.response.status = 403
             return trans.show_error_message("Must be logged in to create workflows")
         if (job_ids is None and dataset_ids is None) or workflow_name is None:
             jobs, warnings = summarize(trans, history)

lib/galaxy/workflow/extract.py

@@ -321,7 +321,15 @@ def __summarize_dataset_collection(self, dataset_collection):
         # tracking with creating_job_associations. Will delete at some point.
         elif dataset_collection.implicit_output_name:
             # TODO: Optimize db call
-            dataset_instance = dataset_collection.collection.dataset_instances[0]
+            element = dataset_collection.collection.first_dataset_element
+            if not element:
+                # Got no dataset instance to walk back up to creating job.
+                # TODO track this via tool request model
+                job = DatasetCollectionCreationJob(dataset_collection)
+                self.jobs[job] = [(None, dataset_collection)]
+                return
+            else:
+                dataset_instance = element.hda
             if not self.__check_state(dataset_instance):
                 # Just checking the state of one instance, don't need more but
                 # makes me wonder if even need this check at all?