Function load_kimi_layer 

pub fn load_kimi_layer(
    shard: &ShardedSafetensors,
    hp: &KimiRealHparams,
    kind: LayerAttentionKind,
    is_dense: bool,
    layer_idx: usize,
) -> Result<KimiDecoderLayerWeights, KimiLoadError>

Loads any one layer (KDA or Gated-MLA attention, dense or MoE FFN), dispatching on kind/is_dense – pass ModelConfig::layer_attention_kind(layer_idx)/ ModelConfig::layer_is_dense(layer_idx) for Kimi K3’s real per-layer topology.