A few thoughts - not a complete answer.
CCD devices tend to have higher power consumption. As you go further to the IR, the bandgap of the device has to be smaller (since the photons have lower energy at those wavelengths) and you want to keep your power consumption down so the device doesn't heat up (which would create noise). Silicon cuts off at around 1100 nm - not far enough for the application you describe.
Strictly speaking, "CCD" refers not to the detection of the light, but to the readout scheme. Each pixel has a corresponding capacitor that stores the integral of the light received, and through a "bucket brigade" type of circuit the charge can be communicated from one pixel to the next until you reach the edge of the sensor, where an ADC read the value.
By contrast, a CMOS sensor (really, CMOS refers to the complementary metal oxide semiconductor process used to place transistors on the semiconductor substrate) is "random access" - that is, you turn on a transistor to read back the voltage from a particular pixel. This has certain advantages (speed of readout of a small region of the sensor, power dissipation) but some disadvantages (noise). But as CMOS technology is very widely used, tremendous advantages of scale and cost, as well as performance, have emerged over time - making CMOS preferable in almost any application.
What you need for IR is low power dissipation, low noise, and small bandgap. That means you need a different material for detecting the light; and possibly a different technology for reading out the pixels.
One of the supplies of IR imaging technology for astronomy is Teledyne, and they published a nice description of some of the technologies they developed in 2008. Reading this, I find that they use HgCdTe, with a variable cutoff wavelength depending on the exact application. The nice thing about MCT (Mercury Cadmium Telluride) is that you can "tune" the bandgap between 0.1 eV and 1.5 eV by changing the ratio of Hg and Cd. By growing the detector using molecular beam epitaxy (MBE), this ratio can be carefully controlled - and thus you can tune a detector to be "just right" for the wavelength of interest. If the bandgap is too large, you won't see the longest wavelength of interest; if it's too small, you will get excess thermal noise. The unique flexibility of MCT makes it the preferred material for high quality IR sensors.
The figure 2 from that paper is worth reproducing here - it shows you the structure of these devices, and shows that the IR detection and readout are two distinct functions, and that the use of different technologies for each helps optimize performance: