Mark full sync data parallel ops with rules

Instead of using hard-coded rules or rely on gpu_strategy to mark full sync data parallel ops, we need some generic rules that is applicable to both the single and distributed setting.

Author: mraway

caffe2/python/layer_model_helper.py

@@ -404,7 +404,8 @@ def output_schema(self, schema):
 
     @property
     def preproc_output_schema(self):
-        assert self._preproc_output_schema is not None
+        if self._preproc_output_schema is None:
+            self._preproc_output_schema = self.input_feature_schema + self.trainer_extra_schema
         return self._preproc_output_schema
 
     @preproc_output_schema.setter