Model-driven context management

Key features

• Model driven - summary written by the model itself, no second LLM summary call (cost advantage of ACP)
• Only suggestions, not mandatory - the automatic strategy only nudges (reminds), whether and when to compress is determined by the model
• Durable and recoverable - the compressed range becomes a checkpoint node and the original text is retained in the append-only session log
• decompress can be restored, search_context can be found within the block
• Based on seq references - no message tags required