Implement base compute metrics

bayesflow/experimental/networks/summary_network.py

@@ -6,4 +6,10 @@
 
 class SummaryNetwork(keras.Layer):
     def compute_metrics(self, data: dict[str, Tensor], stage: str = "training") -> dict[str, Tensor]:
-        raise NotImplementedError
+
+        # TODO - find a good way to deal with summary conditions
+        data["summary_outputs"] = self(
+            data["summary_variables"], training=stage == "training"
+        )
+        # TODO metrics
+        return {}