Calculate approximate context-window usage, remaining token capacity, and output reserve from a model context limit and prompt size.