The ZeroInflatedRegressor class consists of a classifier and a regressor and it says in the documentation that "The regressor is only trained on examples where the target is non-zero, which makes it easier for it to focus." In the code, however, the regressor is actually trained on examples where the classifier-predicted target is non-zero.
The Bug
How to Fix the Bug