fix(mlflow): call model_dump() instead of passing method reference in _add_chunk_events

choice.delta.model_dump (without parens) passes the bound method object to
json.dumps instead of calling it, causing _add_chunk_events to always raise
an exception silently. This prevents _end_span_or_trace from being called,
so zero MLflow traces are recorded for streaming responses.

Fixes: streaming traces not appearing in MLflow experiment tne-costs

litellm/integrations/mlflow.py

@@ -157,7 +157,7 @@ class MlflowLogger(CustomLogger):
                     SpanEvent(
                         name="streaming_chunk",
                         attributes={
-                            "delta": json.dumps(choice.delta.model_dump, default=str)
+                            "delta": json.dumps(choice.delta.model_dump(), default=str)
                         },
                     )
                 )