Update backend API interfaces to be agnostic to messages list (i.e. MessageGraph)

[andrewnguonly]

Summary

Currently, the OpenGPTs backend only implements MessageGraph. Implementing StateGraph in the backend demonstrates the ability to support arbitrary graphs within OpenGPTs. Migrating the current usage of MessageGraph to StateGraph is the first step.

Changes in this PR are adopted from this PR.

Implementation

1. Rename internal variables and update docstrings: messages --> values.
2. Add support for Dict[str, Any] agent state type: Sequence[AnyMessage] --> Union[Sequence[AnyMessage], Dict[str, Any]].
3. Breaking API changes to rename interfaces to be agnostic to a agent state type.

Breaking API Changes

1. GET /threads/<tid>/messages --> GET /threads/<tid>/state
   Response: {"messages": [...], ...} --> {"values": [...], "next": []}
2. POST /threads/<tid>/messages --> POST /threads/<tid>/state
   Body: {"messages": [...]} --> {"values": [...]}

To Do

• Migrate get_chatbot_executor() to use StateGraph
• Migrate get_retrieval_executor() to use StateGraph
• Migrate get_google_agent_executor() to use StateGraph
• Migrate get_openai_agent_executor() to use StateGraph
• Migrate get_xml_agent_executor() to use StateGraph

Note: This PR can be merged without completing these to-do's.

[andrewnguonly]

Update: Chatted with @nfcampos offline and the items in the To Do list will be implemented in a separate PR. get_chatbot_executor() and get_retrieval_executor() will be updated, but the rest will be left alone.