Function moe_hash_route_batched 

pub fn moe_hash_route_batched(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    gate_logits: DevicePtr,
    tid2eid: DevicePtr,
    token_ids: DevicePtr,
    expert_indices: DevicePtr,
    expert_weights: DevicePtr,
    num_experts: u32,
    top_k: u32,
    normalize: bool,
    scaling_factor: f32,
    n: u32,
    stream: u64,
) -> Result<()>

Batched GPU-side MoE hash routing (DeepSeek-V4 hash_moe layers, prefill).

One block per token; reads token_ids[N] and the static tid2eid table. Grid: (N, 1, 1) Block: (256, 1, 1)