The transition to stationary populations is between stage 4 to stage 5 in all exemples of the DT model that I've seen. Not between stage 3 and 4. This is also what is being implied in the figure that you've included. In your link they also write:
Countries like India in the third phase of demographic transition have fertility rates that have declined significantly from previously high levels but have not reached the population-stabilizing "replacement level" of 2.1 children per woman.
This is not saying that countries reach the replacement level 2.1 at the end of stage 3.
The same stages are also described at the Wikipedia page on DT and at populationeducation.org.
So you are correct; at replacement levels, populations shouldn't increase and the net growth should be zero.