I'm happy to agree that none of this should be held up by a dispute over mere semantics. And I agree that languages evolve, and is used different but consistently by different communities at different times. But let me just note that since most scientific journals and some newspapers use "retract" to mean disavowal without textual removal, clearly describing whatever NPR and the NYTimes are doing requires an additional word (or at least a modifier). The term I've heard used sometimes, and which I think is clear, is "unpublish" (but this is not standardized).