Unpin langgraph stack; upgrade langchain-core for Reviver compat

Closes #10

Remove force-pins on langgraph-checkpoint/prebuilt/sdk and bump
langchain-core (and companions) so checkpoint 4.x Reviver(allowed_objects=...)
works. Keep langgraph on a supported range matching langchain 1.3.x.
Also switch BaseMessage .text() calls to the .text property.

llm_be/chat_backend/services/llm_service.py

@@ -126,7 +126,7 @@ class AsyncLLMService(LLMService):
         # return "\n".join(
         #     f"{'User' if prompt.is_user else 'AI'}: {prompt.text}" for prompt in prompts
         # )
-        return "\n".join([f"{"User" if prompt.type=="human" else "AI"}: {prompt.text()}" for prompt in conversation])
+        return "\n".join([f"{"User" if prompt.type=="human" else "AI"}: {prompt.text}" for prompt in conversation])
 
     async def _get_recent_messages(self, conversation: list) -> str:
         """Async version of format conversation history."""
@@ -139,7 +139,7 @@ class AsyncLLMService(LLMService):
         # return "\n".join(
         #     f"{'User' if prompt.is_user else 'AI'}: {prompt.text}" for prompt in prompts
         # )
-        return "\n".join([f"{"User" if prompt.type=="human" else "AI"}: {prompt.text()}" for prompt in conversation])
+        return "\n".join([f"{"User" if prompt.type=="human" else "AI"}: {prompt.text}" for prompt in conversation])
 
     async def generate_response(
         self, conversation: Conversation, query: str, conversation_id: int, **kwargs

llm_be/chat_backend/services/rag_services.py

@@ -330,7 +330,7 @@ class AsyncRAGService(RAGService):
         """Format conversation history for the prompt."""
         return "\n".join(
             [
-                f'{"User" if prompt.type == "human" else "AI"}: {prompt.text()}'
+                f'{"User" if prompt.type == "human" else "AI"}: {prompt.text}'
                 for prompt in conversation
             ]
         )