When user explicitly passes --backend nvidia with a GGUF model,
vulkan is now removed from the available backends list since
llama-cpp-python will use CUDA instead of Vulkan.