[v3-dev] Dedicate one instance of LLM memory per chat

[dlqqq]

Problem

In v3-dev, LLM memory is still a singleton that is shared across all chats. If you ask a question "What is 1+1" in foo.chat, you can still ask questions about your previous question in bar.chat.

Proposed Solution

• Split LLM memory, i.e. dedicate one instance of LLM memory per chat
• LLM memory should call a method on the YChat instance to fetch the most recent messages, instead of maintaining a list in-memory.
  • Add this method if it does not exist.