Function matmul 

pub fn matmul(
    a: &[f32],
    a_rows: usize,
    a_cols: usize,
    b: &[f32],
    b_rows: usize,
    b_cols: usize,
    _mode: SimdMode,
) -> Result<Vec<f32>, EngineError>

C = A @ B^T style? We use row-major: out[m,n] = sum_k a[m,k] * b[k,n] with a: [m,k], b: [k,n].