Optimize gpu reductions (#27)

* Add reduction clause to target_teams_distribute

* Add reductions tests for nested for under parallel

* Optimize GPU reductions
  - Use a 2-level approach with atomics
  - Support DSA_REDUCTION_MUL for nested for directices

* Clean up code

Author: ggeorgakoudis

src/numba/openmp/__init__.py

@@ -7448,7 +7448,7 @@ def NUMBER(self, args):
                                   | thread_limit_clause
                                   | data_default_clause
                                   | data_sharing_clause
-                           //     | reduction_default_only_clause
+                                  | reduction_clause
                                   | lastprivate_clause
                                   | collapse_clause
                                   | dist_schedule_clause