Function mla_q_final_assemble_batched 

pub fn mla_q_final_assemble_batched(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    q_absorbed: DevicePtr,
    q_rope: DevicePtr,
    q_final: DevicePtr,
    num_tokens: u32,
    nq: u32,
    kv_lora: u32,
    rope: u32,
    mla_cache_dim: u32,
    stream: u64,
) -> Result<()>

2026-09-25: Build Q as [q_absorbed (kv_lora) | q_rope (rope)] per head and token, mla_cache_dim = kv_lora + rope wide.