I would assume retiring Theano implies they are working on some other framework. What is it?
I'd say it's rather that the that the features Theano pioneered can now be had from several other well-backed projects. I have no personal insight to MILA, but I suspect the development of Theano must be consuming considerable work time from academic researchers, and they concluded that time and effort would be better spent e.g. on developing and examining ML algorithms.
I wonder if some other organization will pick up responsibility for Theano. Seems to me like a project that might be difficult to keep healthy on only individual volunteer contributions.