A good question and I have already planned to check that approach too. But undersampling means that the noise will sneak into the baseband via aliasing, which in turn implies a need to attenuate the unwanted spectral component even more at the input side. Hence a decent analog filter.
This mixed-mode chip has 4 mixers and each of them can be configured to work in continuous time (multiplying) or sampled mode. Since the LO NCO must be there anyway, it would be strange for the initial approach not to use a component which is already idling on the chip. And from there, the usual 4*BW high resolution quadrature sampling, but at merely 10,4+kHz, not 310kHz.
That would mean adaptive sampling frequency selection based on the spectral components present in the actual deployment environment. Doable, but why, if a passband has the left skirt attenuation too?
Best regards, Piotr