Current pair-conversion event generators, e.g. in Geant4, have been developed mainly for the simulation of electromagnetic showers in calorimeters. They don't sample the five-dimensional differential cross-section (5D DCS), but a product of 1D DCSs. Most of them use high-energy and/or small angle approximations. Also the e+ and e- polar angles are generated independently so energy-momentum is not conserved. None of them can simulate the conversion of polarized photons correctly.
In the past I wrote a generator that is sampling exactly the 5D Bethe-Heitler DCS. I used the VEGAS method: at a given energy, for a given target nucleus, after a 5D grid has been optimized, the DCS is tabulated, something that needs several seconds. Then zillions of conversions can be generated quickly at that energy and for that target. That software has undergone an extensive validation campaign. With an appropriate definition of the event azimuthal angle it was shown that the polarization asymmetry of the conversion agrees nicely with the low- and high-energy known asymptotic expressions.
I have developed a VEGAS-free version that allows the fast generation of the conversion of a photon of a given energy on a given target, with the same other properties as for the VEGAS-based generator.