Function rms_norm_residual 

pub fn rms_norm_residual(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    input: DevicePtr,
    weight: &DenseWeight,
    output: DevicePtr,
    residual: DevicePtr,
    num_tokens: u32,
    hidden_size: u32,
    eps: f32,
    stream: u64,
) -> Result<()>

Fused RMS norm + residual save: normed = rms_norm(input), residual = input.

Eliminates a separate D2D copy by writing the raw input to the residual buffer in the same pass as the normalized output write.

Kernel: rms_norm_residual(input, weight, output, residual, hidden_size, eps) Grid: (num_tokens, 1, 1) Block: (min(hidden_size, 1024), 1, 1)