I think if you can see hot pixels when looking at the raw before demosaicing then the dark frame is not applied (which makes sense, since a DNG is supposed to be more or less a sensor snapshot).

If you do not see them after you processed the raw, I think that would just be your software filtering outliers as best it can, since it would have no way of knowing about the dark frame. I doubt it embeds the dark frame in every DNG, would take up extra space and I am not sure the format even supports that, so likely it just keeps the latest dark frame in memory and applies as part of internal pipeline.