BUG: LiteLLMModel crashes when multiple consecutive system messages are provided #1972

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
Cell In[2], line 15
      4 model = LiteLLMModel(  
      5     model_id="gemini/gemini-2.5-flash", 
      6     api_key=os.getenv("GOOGLE_API_KEY"),  
      7 )  
      9 messages = [  
     10     {"role": "system", "content": "When you say anything Start with 'FOO'"},  
     11     {"role": "system", "content": "When you say anything End with 'BAR'"},  
     12     {"role": "user", "content": "Just say '.'"}  
     13 ]  
---> 15 response = model(messages)  
     16 print(response.content)

File /opt/homebrew/Caskroom/miniconda/base/envs/smolagents/lib/python3.12/site-packages/smolagents/models.py:581, in Model.__call__(self, *args, **kwargs)
    580 def __call__(self, *args, **kwargs):
--> 581     return self.generate(*args, **kwargs)

File /opt/homebrew/Caskroom/miniconda/base/envs/smolagents/lib/python3.12/site-packages/smolagents/models.py:1274, in LiteLLMModel.generate(self, messages, stop_sequences, response_format, tools_to_call_from, **kwargs)
   1266 def generate(
   1267     self,
   1268     messages: list[ChatMessage | dict],
   (...)   1272     **kwargs,
   1273 ) -> ChatMessage:
-> 1274     completion_kwargs = self._prepare_completion_kwargs(
   1275         messages=messages,
   1276         stop_sequences=stop_sequences,
   1277         response_format=response_format,
   1278         tools_to_call_from=tools_to_call_from,
   1279         model=self.model_id,
   1280         api_base=self.api_base,
   1281         api_key=self.api_key,
   1282         convert_images_to_image_urls=True,
   1283         custom_role_conversions=self.custom_role_conversions,
   1284         **kwargs,
   1285     )
   1286     self._apply_rate_limit()
   1287     response = self.retryer(self.client.completion, **completion_kwargs)

File /opt/homebrew/Caskroom/miniconda/base/envs/smolagents/lib/python3.12/site-packages/smolagents/models.py:523, in Model._prepare_completion_kwargs(self, messages, stop_sequences, response_format, tools_to_call_from, custom_role_conversions, convert_images_to_image_urls, tool_choice, **kwargs)
    521 # Clean and standardize the message list
    522 flatten_messages_as_text = kwargs.pop("flatten_messages_as_text", self.flatten_messages_as_text)
--> 523 messages_as_dicts = get_clean_message_list(
    524     messages,
    525     role_conversions=custom_role_conversions or tool_role_conversions,
    526     convert_images_to_image_urls=convert_images_to_image_urls,
    527     flatten_messages_as_text=flatten_messages_as_text,
    528 )
    529 # Start with messages
    530 completion_kwargs = {
    531     "messages": messages_as_dicts,
    532 }

File /opt/homebrew/Caskroom/miniconda/base/envs/smolagents/lib/python3.12/site-packages/smolagents/models.py:376, in get_clean_message_list(message_list, role_conversions, convert_images_to_image_urls, flatten_messages_as_text)
    373                 element["image"] = encode_image_base64(element["image"])
    375 if len(output_message_list) > 0 and message.role == output_message_list[-1]["role"]:
--> 376     assert isinstance(message.content, list), "Error: wrong content:" + str(message.content)
    377     if flatten_messages_as_text:
    378         output_message_list[-1]["content"] += "\n" + message.content[0]["text"]

AssertionError: Error: wrong content:When you say anything End with 'BAR'