"Canonical" refers largely to the names we give to things, so that we maximize the extent to which an isomorphism maps the object named A in X to the object named A in Y. And also to choosing an isomorphism that embeds in a different isomorphism (of superstructures) that isn't strictly the topic of the current question.
I don't think anyone thinks canonical isomorphisms are mathematically controversial (except some people having fun with studying scenarios where more than one isomorphism is equally canonical, and other meta studies), they are a convenient communication shorthand for avoiding boring details.