Function launch_k3_latent_moe_experts 

pub fn launch_k3_latent_moe_experts(
    gpu: &dyn GpuBackend,
    kernels: &K3MoeGemmKernels,
    packed: &[(String, QuantizedWeight)],
    latent: &[f32],
    ids: &[usize],
    mix_w: &[f32],
    cfg: &LatentMoeConfig,
    stream: u64,
) -> Result<Vec<f32>>

2026-09-25: Computes what mix_routed_experts computes, sum_e mix_w[e] * w2(SiTU(w1 x, w3 x)) over ids, from the packed experts, with BF16 GEMM inputs and outputs. Empty ids returns zeros without a launch.