fix(autogpt/llm): AssistantChatMessage.tool_calls default [] instead of None

OpenAI ChatCompletion calls fail when `tool_calls = None`. This issue came to light after 22aba6d.

autogpts/autogpt/autogpt/core/resource/model_providers/openai.py

@@ -423,7 +423,7 @@ class OpenAIProvider(
                 tool_calls=(
                     [AssistantToolCall(**tc.dict()) for tc in _assistant_msg.tool_calls]
                     if _assistant_msg.tool_calls
-                    else None
+                    else list()
                 ),
             )
             response = ChatModelResponse(

autogpts/autogpt/autogpt/core/resource/model_providers/schema.py

@@ -44,16 +44,14 @@ class ChatMessage(BaseModel):
         SYSTEM = "system"
         ASSISTANT = "assistant"
 
+        TOOL = "tool"
+        """May be used for the result of tool calls"""
         FUNCTION = "function"
         """May be used for the return value of function calls"""
 
     role: Role
     content: str
 
-    @staticmethod
-    def assistant(content: str) -> "ChatMessage":
-        return ChatMessage(role=ChatMessage.Role.ASSISTANT, content=content)
-
     @staticmethod
     def user(content: str) -> "ChatMessage":
         return ChatMessage(role=ChatMessage.Role.USER, content=content)
@@ -93,7 +91,7 @@ class AssistantToolCallDict(TypedDict):
 class AssistantChatMessage(ChatMessage):
     role: Literal["assistant"] = "assistant"
     content: Optional[str]
-    tool_calls: Optional[list[AssistantToolCall]]
+    tool_calls: list[AssistantToolCall] = Field(default_factory=list)
 
 
 class AssistantChatMessageDict(TypedDict, total=False):