Function mla_q_rope_scatter 

pub fn mla_q_rope_scatter(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    q_full: DevicePtr,
    q_absorbed_buf: DevicePtr,
    q_rope_contiguous: DevicePtr,
    nq: u32,
    hd: u32,
    nope: u32,
    rope: u32,
    kv_lora: u32,
    mla_cache_dim: u32,
    stream: u64,
) -> Result<()>

2026-09-25: Copy each head’s rope part of q_full ([nq, hd], at offset nope) into q_absorbed_buf ([nq, mla_cache_dim], at offset kv_lora) and into the contiguous q_rope_contiguous ([nq, rope]) in one pass.