problem is your interviewer (and also this article) wrongly assumed logistic regression is always calibrated because he/she was reading scikit learn documentation.

In reality logistic regression is neither calibrated by default nor always well calibrated in practice. There are papers showing theoretically that logistic regression is miscalibrated and it is also very easy to observe it in practice.

https://github.com/valeman/calibration_is_what_you_need