Function w8a8_prefill_kernels_loaded 

pub fn w8a8_prefill_kernels_loaded(gpu: &dyn GpuBackend) -> bool

Whether BOTH W8A8_PREFILL_KERNELS are loaded for this target, asked of the backend rather than of a constructed layer.

Same answer Qwen3AttentionLayer::has_w8a8_prefill_kernels gives — the layer just caches the handles init.rs already resolved through try_kernel, and try_kernel returns KernelHandle(0) for an absent kernel exactly as this does.