It is by design. Built in features like this one caused some trouble for people, and not everyone had much use for it, so I decided to make the system easier to use and add minimal built-in features so that anyone who needs a certain feature can easily extend the system without much code-reworking.
Basically, what you need to do is add a new node, with a function connected to a 'started new message' signal. In that function you will get the Dialogue.current_speaker_node and then its face texture value (you need to define it in the speaker script first, of course). And then just set the texture to the BoxMessage/TextBackground/_hbox/Face node.