Fix the performance of the database query in the user_results tag

The commit a6c1ea7 tried to optimize
the submitter statistics in the results page (Your points). However,
the database query in the `user_results` template tag was too slow
on courses with hundreds of students and thousands of submissions.

The order of `filter()` and `annotate()` affects the results, that is,
they are not commutative. The new version in this commit seems to
work correctly and performs quite well.

https://docs.djangoproject.com/en/3.2/topics/db/aggregation/#order-of-annotate-and-filter-clauses

exercise/templatetags/exercise.py

@@ -79,9 +79,9 @@ def user_results(context: Context, student: Optional[User] = None) -> Dict[str,
         instance = context['instance']
         values['student_count'] = instance.students.count()
         counts = (instance.students
+            .filter(submissions__exercise__course_module__course_instance=instance)
             .values('submissions__exercise_id')
             .annotate(count=models.Count('submissions__submitters', distinct=True))
-            .filter(submissions__exercise__course_module__course_instance=instance)
             .order_by()
         )
         values['exercise_submitter_counts'] = {row['submissions__exercise_id']: row['count'] for row in counts}