Function paged_decode_attn_reduce_nvfp4 

pub fn paged_decode_attn_reduce_nvfp4(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    workspace: DevicePtr,
    output: DevicePtr,
    seq_lens: DevicePtr,
    num_q_heads: u32,
    head_dim: u32,
    num_splits: u32,
    num_seqs: u32,
    stream: u64,
) -> Result<()>

2026-09-25: Combine the split-K partials in workspace into the BF16 output.