pub fn hc_pre( x: &[f32], pre: &[f32], tokens: usize, hc: usize, dim: usize, ) -> Vec<f32>
Block.hc_pre: collapse the copies, y[d] = sum_c pre[c] * x[c][d], back to bf16.
Block.hc_pre
y[d] = sum_c pre[c] * x[c][d]