Function bessel_jv_impl 

pub fn bessel_jv_impl<P, E, V, const A1: usize, const A2: usize, const AH: usize, const B1: usize, const B2: usize, const BH: usize>(
    x: V,
    n: V,
    t0: &BesselJ<E, A1, A2, AH>,
    t1: &BesselJ<E, B1, B2, BH>,
) -> V
where
    E: FloatElement,
    V: FloatVector<Element = E> + TranscendentalMathWithPolicy,
    P: Policy,

J_n(x) with a per-lane order.

Both arms, selected per lane as in the const form: forward where n < x, downward ratio otherwise. The forward arm freezes each lane at its own order. The downward arm was already masked on k <= N, so a vector n slots straight in.

The J_0-zero guard carries over unchanged and matters just as much: r_1 is kept out of the running product and folded into the J_0 seed, so a lane sitting on a zero of J_0 takes the J_1 seed without ever forming inf * 0.