Function forward_token 

pub fn forward_token(
    model: &K3CpuModel,
    token: u32,
    pos: usize,
    cache: &mut HybridCache,
    ablation: Ablation,
) -> Vec<f32>

Embed one token and run every decoder layer at pos. Updates the hybrid cache. AttnRes is per-token (layer depth), not carried across the sequence.