What technique is used to determine the energies of the radionuclide standards used in energy sensor calibration?

If you really want a historical answer, I don't know for sure. However, these detectors are highly linear, so all you really need is two calibration points for a particular detector, with a particular amount of high voltage applied to it. For example, if you have a beta+ source and a NaI detector (a fairly old technology), then what you're going to see in the detector's spectrum is a photopeak with a known energy of 511 keV, and also a Compton edge which has a known energy. For a cleaner calibration, maybe build your detector with a cavity inside and insert the source into the cavity. Then you'll get events in which both decay gammas are detected, so there will be a second gaussian peak at 2x511 keV.

If you're concerned about the linearity of the detector, then you can study a source that has gammas forming a sum $E_1=E_2+E_3$. It's common these days with HPGe detectors to do a second-order polynomial fit with a set of known sources. But in the early days when they needed to get started with known standards, I suspect the nonlinearity was negligible compared to the lousy energy resolution of a NaI.

At the low energy end, you may also be able to compare with diffraction or take measurements with theoretically calculable K-shell energies. I don't know.