We report deep EVN and eMERLIN observations of the Type Ia SN 2014J in the nearby galaxy M 82. Our observations represent, together with JVLA observations of SNe 2011fe and 2014J, the most sensitive radio studies of Type Ia SNe ever. By combining data and a proper modeling of the radio emission, we constrain the mass-loss rate from the progenitor system of SN 2014J to $dot{M} lesssim 7.0times 10^{-10}, {rm M_{odot}, yr^{-1}}$ (3-$sigma$; for a wind speed of $100, {rm km s^{-1}}$). If the medium around the supernova is uniform, then $n_{rm ISM} lesssim 1.3 {rm cm^3}$ (3-$sigma$), which is the most stringent limit for the (uniform) density around a Type Ia SN. Our deep upper limits favor a double-degenerate (DD) scenario--involving two WD stars--for the progenitor system of SN 2014J, as such systems have less circumstellar gas than our upper limits. By contrast, most single-degenerate (SD) scenarios, i.e., the wide family of progenitor systems where a red giant, main-sequence, or sub-giant star donates mass to a exploding WD, are ruled out by our observations. Our estimates on the limits to the gas density surrounding SN 2011fe, using the flux density limits from Chomiuk et al. (2012), agree well with their results. Although we discuss possibilities for a SD scenario to pass observational tests, as well as uncertainties in the modeling of the radio emission, the evidence from SNe 2011fe and 2014J points in the direction of a DD scenario for both.