fixed boundary case in supply and demand

We protect against the boundary case of having no choosers in the predict method, but this gets circumvented and is still a problem with the supply and demand code.

I was hoping to protect against this outside of urbansim, but we dont't know how many choosers there are until the predict_filters gets run, and that happens in urbansim.

urbansim/models/dcm.py

@@ -1135,6 +1135,9 @@ def summed_probabilities(self, choosers, alternatives):
             Summed probabilities from each segment added together.
 
         """
+        if len(alternatives) == 0 or len(choosers) == 0:
+            return pd.Series()
+
         logger.debug(
             'start: calculate summed probabilities in LCM group {}'.format(
                 self.name))