Well, I see what you are saying, and you obviously understand about Nyquist and all that, so I have no quarrel with you.
But to me it is not clear that the OP can undersample. If the Doppler return is from a complicated scene with multiple targets, or rotating targets (which therefore have a large range of Doppler) and all of this information needs to be preserved, then undersampling is probably not OK.
If there is a single dominant frequency, and the OP just wants to measure what the frequency is, then it may be enough to use some kind of frequency counter scheme instead of using an ADC.
--Mac