One model to enhance them all: array geometry agnostic multi-channel personalized speech enhancement.