Why is the logic of the MultiAgentFormatter different between the openai and other model?

Java version uses `appendRoleAndName` while Python version only appends the name: `accumulated_text.append(f"{msg.name}: {block['text']}")`