That basic idea is old as the hills
> there's the 1:1 requirement.
In most cases, the DUT has gain, so you put a resistive divider on the output and trim for overall unity gain. That means a precision diff-amp is NOT needed, what you really end up with is a diff-amp with different and adjustable gains on each input. You need high internal CMRR, and the classic 4-R diff-amp may not be as good as an inverter-summer.
Also note that if the DUT is inverting, you don't need no diff-amp. Just sum the input and output with adjustable gain to get a null.
> any phase lag through the device-under-test will (wrongly) show up as distortion.
That is sure true in the classic use, with voltmeter readout. For midband audio, a very small phase-shift will give a null on the fundamental (though it makes the harmonics a mess).
But what is novel here is driving a spectrum analyzer. So I don't think fundamental phase matters. It isn't vital to get a DEEP null. What they are doing is getting "most" of the fundamental out of the result, so the spectrum analyzer input isn't stuffed full of strong fundamental while trying to read a weak residual.