Make the AverageLearner only return new points ...

and introduce learner.pending_points to make it more similar to the rest of Adaptive.

I've added a test that will fail with the current master.

adaptive/learner/average_learner.py

@@ -110,7 +110,7 @@ class AverageLearner(BaseLearner):
 
     def remove_unfinished(self):
         """Remove uncomputed data from the learner."""
-        pass
+        self.pending_points = set()
 
     def plot(self):
         hv = ensure_holoviews()