Sort DF by probability when loading from parquet

src/av2/datasets/motion_forecasting/eval/submission.py

@@ -88,6 +88,7 @@ def from_parquet(cls, submission_file_path: Path) -> ChallengeSubmission:
         """
         # Load submission data and sort rows by descending probability
         submission_df = pd.read_parquet(submission_file_path)
+        submission_df.sort_values(by="probability", inplace=True, ascending=False)
 
         # From serialized data, build scenario-track mapping for predictions
         submission_dict: Dict[str, ScenarioPredictions] = defaultdict(lambda: defaultdict(dict))  # type: ignore