fix: DEFAULT FEATURES must be present when creating metadata. If not, then we raise. this is a first step towards standardazing the dataset format, or otherwise (as it is now) everything would be allowed

src/lerobot/datasets/lerobot_dataset.py

@@ -512,7 +512,9 @@ class LeRobotDatasetMetadata:
 
         obj.root.mkdir(parents=True, exist_ok=False)
 
-        features = {**features, **DEFAULT_FEATURES}
+        assert set(DEFAULT_FEATURES.keys()).issubset(set(features.keys())), (
+            f"Default features must be a subset of the provided features! Got {set(features.keys())} and {set(DEFAULT_FEATURES.keys())}"
+        )
         _validate_feature_names(features)
 
         obj.tasks = None