First, note that the circuit construction using classical condition like you have is not executable on IBM hardware at the moment. Devices like Honewywell Trapped ion allows you to do such thing (I think). However, thanks to principle of deferred measurement, we can push the measurement all the way back to the end of the circuit. See here.
Essentially, your circuit becomes

The state of $q2$ will depends on the state of $q0$. If $q0$ starts in the state $|0\rangle$ then measurement in $q2$ will always (assume no noise) be 0. If instead I add the $X$ operation on $q0$ to turn it into the state $|1\rangle$ then measurement on $q2$ will registered a 1.
For example: if I initialize $q0$ to the state $|0\rangle$ then I will have

but if I initialize $q0$ to the state $|1\rangle$ then I will have
