[eluser]TheFuzzy0ne[/eluser]
I see what you're saying, but to my understanding libraries are made to be as self contained as possible, and it's to save code. Using the pagination class as-is, there are no assumptions made about what code you have in place elsewhere (such as in your model). By working with the result offset rather than the page number, it's ready to use as it is, there's no logic that actually has to be added to your models to work out the actual offset, as some people might not understand how the calculations work.
However, with that said, I see no reason why it can't still use the actual page number from the URL, but return an offset to the model. Perhaps it would have been better this way, but then again, perhaps it confuses the issue. The pagination class would still need to know how many results there are, and how many results per page should be displayed. I think that in short, it saves developers having to do the calculations themselves.
Just out of interest, how would you have done this, assuming the user isn't wanting to add any logic to their model to calculate the offset from a page number? If we can agree on a better way, perhaps this could be implemented in the future. This is by no means the first instance I've seen where someone has asked this question.