How does all this work with targets, where seeking is mapped to a long
press of the skip button, like on iPods? Do I have to press the button
for a second for every step? What happens if I hold it?

The current way (seeking with acceleration as long as the button is
being held) seems fine to me, and could only be outperformed by an
apple-like approach of using the wheel for setting the position, which I
would really appreciate in Rockbox.