In the PS output, expose the scheduler selected size (clamped by model context size) instead of always reporting the model max context. This will help provide a hint to clients to keep the context size below this value to avoid paging and poor performance on smaller VRAM systems.
