improve message data model

[mieslep]

https://github.com/langchain-ai/langchain-datastax/blob/main/libs/astradb/langchain_astradb/chat_message_histories.py

creates a new Document per message, and searches are based on

            filter={
                "session_id": self.session_id,
            },

However, this not terribly efficient once we start getting into 10s or 100s of messages in history: many documents need to be returned and then ordered.

Alternate data models could include:

1. Maintaining a list of Message within a single document where the key is the conversation ID
2. Use CassandraChatMessageHistory https://python.langchain.com/docs/integrations/memory/cassandra_chat_message_history/ which stores within a standard Cassandra table that has the conversation ID as the partition key; the Cassandra libraries do have some intelligence around maintaining connections so it does not require a new connection per interaction, and the library correctly supports token/endpoint connection paradigm.