llama.cpp: Server Speculation Fix After Image Processing

llama.cpp released an update fixing a server-side speculation bug that occurred after image processing. This update ensures the actual position, rather than the token count, is passed to drafters after an image, improving speculation accuracy across all drafters.