It’s not entirely clear that a black circle means “yes” or selected, while a white circle means “no” or non-selected. Depending on what the user regards as foreground and background, it may go either way. Consider this (rather contrived) example:
Which one is selected? The one that “lit up” like a light? Or the one that is “filled with ink”?
There is an implicit assumption that being selected should include a graphic addition of something. Thus the addition of a circle within the circle. It may also be thinking by analogy with checkboxes, where you add a check to the square to show selection.
This “works” even when we break the standard and reverse the colors:
This implies it also works when users don’t know the colors, something which matters especially today when we are not confined to grayscale. For this reason, in addition to it just being non-standard, I would always avoid fully filled radio buttons.
Personally, I think the term “radio buttons” refers metaphorically to the behavior (that it’s one-of-many selection), not to the appearance. Other historical factors may have determined the general appearance. At the time GUIs were being invented in the 1970s and 1980s, the most common radio buttons were the presets in car radios. While I’m sure you can find an exception (I have), these generally were (1) rectangular, not circular, and (2) did not indicate a selected state –once pressed, they popped back out, and did not show a mechanical flag or light. So I don’t think skeuomorphism had anything to do with it.