Use choice to choose the 1dim indices into the array, then index it.
In the example you provided, only the number of possible choices affects the nature of the choice, not the actual values (0, 255). Choosing indices is the 1dim problem choice knows how to handle.
choices = numpy.array([[0,0,0],[255,255,255]])
idx = numpy.random.choice(len(choices),4)
choices[idx]