[Feature Request] Add iterative precomputation of compacting buffer during conversation

Bug Description
Please offer an option to "iteratively precompute the compacting buffer" based on the flow of the conversation -- this alleviates the annoyance of having to "plan to compact" as you are rolling through the work with the model and done relatively frequently, would probably be better information than the actual transcript itself (a minimap of the conversation of sorts).

Environment Info

• Platform: darwin
• Terminal: ghostty
• Version: 2.1.81
• Feedback ID: 8c1105f4-9e20-4e18-ac87-f29b67875f08

Errors

[]