Let's discuss the two extremes separately.
(a) maximum diffraction-picture contrast
Then you have the setup as in the following figure from Dopfer's PhD thesis:
In the thesis Dopfer explains the result a bit different in an equivalent gedankenexperiment description, but I find it simpler to describe the real experiment.
From the LiIO3 crystal you get parametric-down converted entangled photons. What's used in this experiment is only the entanglement of the photon momenta.
Measured are only such entangled pairs, i.e., the photons at the detectors D1 and D2 get only registered when they arrive in coincidence such that you are sure to always measure only photons belonging to an entangled pairs.
With D1 in the focal plane registering photons at a specific position (i.e., making this detector sufficiently small) you select photons of a specific momentum (direction, the magnitude is anyway fixed from the phase-matching condition of the down-conversion process), because in the focal plane there's a one-to-one map between the point of photon detection and the momentum direction of the photon, i.e., you prepare (almost ideally) a plane wave.
For the so registered plane wave the partner photon registered at D2 went through the double slit and is also described by an (almost ideal) plane wave with the corresponding momentum and thus you get a double-slit interference pattern with full contrast when detecting many photons with D2 as a function of the position of D2. Since you have "prepared" a plane wave there's no way (not even in principle!) to gain which-way information, i.e., through which slit the photon registered by D2 came.
Shifting D1 within the focal plane shifts the double-slit interference pattern accordingly.
(b) Allowing for which-way information
Now D1 is put in the imaging plane of the Heisenberg lense.
Now photons with any momentum direction are registered. What D1 sees is an image of the illuminated LiIO3 crystal or a smaller part of it depending on the size of D1. Since now the photons registered with D2 correspond to a light with photon momenta in any direction this corresponds to illuminating the Double slit with incoherent light, i.e., for a large detector D1 you get the incoherent superposition of the two single slits. If the slits are small enough you get a single-slit interference pattern from both slits and just add the intensities of these two. If D1 is small you just get the single-slit interference pattern. In any case since in principle you can know the precise position from which slit each photon registered by D2 came there is no two-slit interference pattern anymore.