Hey! I am also in process of re-creating the volatility model. I did it a couple years ago using BUGS in R but as I am shifting to python/pymc3 I wanted to give it a go as well. This thread has been very helpful, so I will have a go and post here.
I understand close to nothing re theano.scan() but @ricardoV94 made this amazing tutorial on fitting iterative RL model which I think can be adopted to the volatility model - in case you are still stuck @dddd1007