Function paged_decode_attn_splitk_nvfp4 

pub fn paged_decode_attn_splitk_nvfp4(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    q: DevicePtr,
    k_cache: DevicePtr,
    v_cache: DevicePtr,
    workspace: DevicePtr,
    block_tables: DevicePtr,
    seq_lens: DevicePtr,
    max_blocks_per_seq: u32,
    num_q_heads: u32,
    num_kv_heads: u32,
    head_dim: u32,
    block_size: u32,
    inv_sqrt_d: f32,
    num_splits: u32,
    q_stride: u32,
    block_stride_bytes: u64,
    data_section_bytes: u64,
    num_seqs: u32,
    stream: u64,
) -> Result<()>

Split-K paged decode attention (NVFP4 KV cache).

Partitions the KV sequence across num_splits CTAs per (q_head, seq). Each CTA computes partial softmax + weighted output, written to workspace.

Grid: (num_q_heads, num_splits, num_seqs) Block: (256, 1, 1)