Function trigamma_impl 

pub fn trigamma_impl<P, E, V>(x_in: V, t: &Trigamma<E>) -> V
where
    P: Policy,
    E: FloatElementWithBits,
    V: FloatVectorWithBits<Element = E> + SpecializedSpecialMath<E>,

Shared trigamma (psi_1) implementation for all real element types.

psi_1(x) = d/dx psi(x), the second derivative of ln Gamma. Structurally much cheaper than digamma: the reduction to the fitted range is a single step rather than a masked walk, because the [1, 2] rational and the x > 4 rational between them already cover everything from 1 upward.

The poles at zero and the negative integers evaluate to +inf, which is the correct two-sided limit (psi_1 has a double pole there, so unlike psi the two one-sided limits agree) and falls out of the reflection term for free.