Function deinterleave_qg_split_qnorm 

pub fn deinterleave_qg_split_qnorm(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    data: DevicePtr,
    q_out: DevicePtr,
    q_norm_weight: DevicePtr,
    num_tokens: u32,
    num_heads: u32,
    head_dim: u32,
    stride: u32,
    eps: f32,
    stream: u64,
) -> Result<()>

Fused deinterleave Q/Gate + per-head Q RMS norm.

Combines deinterleave_qg_split + Q RMS norm into a single kernel, eliminating one global memory round-trip for Q data. Gate is deinterleaved to data[q_total..], Q is deinterleaved → normalized → q_out.