Make conversation management more powerful

In response to this paper: https://arxiv.org/abs/2505.06120

Our experiments confirm that all the top open- and closed-weight LLMs we test exhibit significantly lower performance in multi-turn conversations than single-turn, with an average drop of 39% across six generation tasks. Analysis of 200,000+ simulated conversations decomposes the performance degradation into two components: a minor loss in aptitude and a significant increase in unreliability. We find that LLMs often make assumptions in early turns and prematurely attempt to generate final solutions, on which they overly rely. In simpler terms, we discover that when LLMs take a wrong turn in a conversation, they get lost and do not recover.

We should have powerful tools to:

• clean up long-running conversations
• fork conversations at a certain checkpoint into new conversations
• summarize conversations as a starting point for a new conversation
• edit/remove any existing messages in conversations

No files, tests, or entry points are identified in the issue. First map the existing conversation implementation in this Rust CLI, then define scope and acceptance criteria for cleanup, forking, summarization, and message editing before attempting the work.