Function masked_mat_vec_mul_lazy 

pub fn masked_mat_vec_mul_lazy(
    rho: &[u8; 32],
    y_hat_m: &[MaskedPoly],
    k: usize,
    l: usize,
    out: &mut [MaskedPoly],
)

Low-memory variant of masked_mat_vec_mul: recomputes each a_hat[i][j] polynomial on-the-fly from the public seed rho via SHAKE128 instead of holding the full k×l matrix in memory.

Trade-off identical to mat_vec_mul_lazy (dsa.rs): saves up to 57 KB of stack for the a_hat matrix at the cost of repeated SHAKE128 invocations. Called from sign_internal when both sca-masked-y and low-mem are enabled.