Cover Edge Case: Empty array instead of None for Assistant tool calls

[Eliotdoesprogramming]

Hey didn't see any contributing guidelines so let me know if I can clean up this PR more

Adding an empty array check for assistant message normalization to account for a bug I had encountered working with langchain AIMessage default behavior (but also could be a valid edge case working purely with mistral).

When passing an assistant message to normalizer, if the assistant message has [] as tool calls instead of None, the content will not be added to the message. Having a message with content = None causes errors on chat serving endpoints downstream e.g. vLLM

Current Behavior:

# relevant snippet
            if message.tool_calls is not None:
                for tool_call in message.tool_calls:
                    normalized_tool_call = self._normalize_tool_call(tool_call)
                    tool_calls.append(normalized_tool_call)
            elif message.content:
                aggregated_content.append(self._aggregate_content_chunks(message.content))

# test code
      message = AssistantMessage(role="assistant", content="Hello", tool_calls=[])
      normalized_message = normalizer._aggregate_assistant_messages([message])
      assert normalized_message.content == "Hello" # message content will be None, and assertion fails

If [] is passed to this function, the if statement will evaluate as True, making it so that the elif statement is skipped and any message content is ignored.

Proposed behavior

# relevant snippet
            if message.tool_calls: # empty array evaluates to false
                for tool_call in message.tool_calls:
                    normalized_tool_call = self._normalize_tool_call(tool_call)
                    tool_calls.append(normalized_tool_call)
            elif message.content:
                aggregated_content.append(self._aggregate_content_chunks(message.content))

If [] is passed to this function, the if statement will evaluate as False, making it so that the elif statement will add its message content.

# test code
      message = AssistantMessage(role="assistant", content="Hello", tool_calls=[])
      normalized_message = normalizer._aggregate_assistant_messages([message])
      assert normalized_message.content == "Hello" # message content Hello and assertion fails

Hopefully this helps keep our AI ecosystem bug free 👍