Just how light-sensitive could the non-theoretical cameras be in real-time?
The theoretical limit is detecting individual photons of course. Modern CCDs come quite close to that, and BTW, so do the rods in the human eye.
So, at that level it comes down to how many photons you can catch. A bigger lens or mirror will help(*), and a longer exposure time will help too.
You said "real-time". Does this mean "25 frames per second", so "exposure time 1/25 second"?
That leaves one variable: the size of the camera. How big do you want it to be?
Changing ISO level just changes an amplification factor. A multiplication factor where a single electron of extra sensor charge is just guaranteed to result in an increased (digital) output value is the maximum useful, but remember that an ordinary 8-bit sensor will be saturated at 256 electrons(**) at this level, and you need to detect several thousands of photons(**) if you don't want a really noisy image. So, at really high ISO levels your picture will be either noisy or saturated.
(*) Assuming there are no diaphragms in the optical path. Using diaphragms doesn't make real sense anyway when photographing galaxies. Just focus on infinitive and open the diaphragm to maximum.
(**) I sometimes say photons and sometimes electrons, deliberately. In digital image sensors, capturing of photons results in moving electrons through a potential barrier. in the ideal case, capturing one photon results in a charge build-up of one electron.