Function moe_expert_gate_up_shared_fp8_t 

pub fn moe_expert_gate_up_shared_fp8_t(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    input: DevicePtr,
    gate_weight_t_ptrs: DevicePtr,
    gate_block_scale_t_ptrs: DevicePtr,
    gate_out: DevicePtr,
    up_weight_t_ptrs: DevicePtr,
    up_block_scale_t_ptrs: DevicePtr,
    up_out: DevicePtr,
    expert_indices: DevicePtr,
    sh_gate_t: &Fp8Weight,
    sh_gate_out: DevicePtr,
    sh_up_t: &Fp8Weight,
    sh_up_out: DevicePtr,
    n: u32,
    k: u32,
    top_k: u32,
    stream: u64,
) -> Result<()>

FP8 fused gate+up GEMV (transposed weight). Single-token decode.