On a rangefinder lenses have a short distance to the sensor (called register distance) That means that wideangle lenses will produce extremely acute incidence angles of the light towards the edges and corners of the sensor. Leica and Kodak have managed to get the acceptance angle of the sensor corrected by shifting the microlenses over the sensor towards the edges. But long lenses will produce light rays that are more or less coming in straight. That means the sensor pixels need to have quite a wide acceptance angle for this technique to work. CCD sensors provide this angle. However, Cmos sensors have electronic circuitry between and partly in front of the pixel, reducing the acceptance angle and thus making them unsuitable for this application. This is quite apart from the fact that the image quality of CCD sensors at lower ISO is superior to CMos sensors. It may be that back-lit CMos sensors can solve the dilemma, but they are not (yet?) available for full-frame and even then they will have to prove themselves.
The reason CMos sensors need these electronics is because they are extermely noisy to the extent that the noise cannot be reduced on output,so the pixel levels need to be balanced individually on the sensor.
Actually it's the other way round. Every sensor needs readout electronics. The motivation for having CMOS sensors in the first place is that if you have the readout electronics on the sensor itself, you don't need separate readout electronics as with a CCD. As a result, you have everything on one die and the whole thing becomes much more integrated and cheaper.
The tradeoff for that is that you have to sacrifice some sensor area for per-pixel readout electronics that in a CCD sensor would be useful for potential wells, i.e. pixel area. That's why on a CMOS sensor you have less sensitivity per-pixel and greater variation between individual pixels.
Early CMOS sensors therefore sacrificed a lot of sensor area to readout electronics instead of photosites. This earned CMOS sensors their early reputation of low sensitivity and lots of noise. As miniaturization advanced, however, the size of the few transistors you need for the readout electronics has been decreasing in relation to the pixel size. So the relative pixel size advantage of CCD sensors is becoming less and less relevant. This is the primary reason why in the past we've been able to observe enormous progress in CMOS sensors, as witnessed in the outstanding lowlight capabilities of cameras like the recent Nikons.
The electronics themselves can be built rather flat in comparison to the size of the sensor cells. As far as I can see, the problem is not primarily that the 3D structure somehow obstructs light rays at extremely oblique angles, but that for an on-sensor microlens array it helps if the underlying pixels are large. I don't think there's anything that precludes CMOS sensors in digital rangefinder cameras.
Leica solved this problem by using a (slightly dated) Kodak CCD, but they could have solved it by using a halfway modern large-photosite CMOS sensor as well if they had got access to one. I guess they didn't find a supplier, or they didn't have enough free electronics R&D capacity to redesign their in-camera electronics for a CMOS sensor, or both.
