Function quantized_any 

pub fn quantized_any(
    store: &WeightStore,
    prefix: &str,
    n: usize,
    k: usize,
    gpu: &dyn GpuBackend,
    variant: Nvfp4Variant,
    qctx: QuantizeCtx,
) -> Result<QuantizedWeight>

2026-09-25: Load a weight [n, k] as NVFP4 whatever its on-disk format.

The variant is overridden per key: a key with only a .weight is quantized as Bf16Raw, and an FP8E4M3 key with a weight_scale or weight_scale_inv but no NVFP4 global scale is dequantized as Fp8Dequanted (unless variant is already Fp8Dequanted or Bf16Raw). n and k are used only by those two re-quantizing paths.