Give the the test_xcom_ar_map test opportunity to flush the data

Using same session in different steps of the same test has the
potential of not flishing/committing the changes between.

Seems that the apache#33150 traded one flakiness with another. Attempting to
make sure that the flash/commit is executed before the second run.

Error:

The test RuntimeError: number of values in row (0) differ from number of
column processors (29)

The error is strange however and indicates a bug in sqlite library.

tests/models/test_xcom_arg_map.py

@@ -253,6 +253,9 @@ def pull(value):
     for ti in decision.schedulable_tis:
         ti.run(session=session)
 
+    session.flush()
+    session.commit()
+
     # Now "pull" should apply the mapping functions in order.
     decision = dr.task_instance_scheduling_decisions(session=session)
     for ti in decision.schedulable_tis: