The mysteries of the Amazon KDP process are mysterious.
I can say that the samples I downloaded didn't have text start offsets in them, but the (mobi7) preview that you downloaded from your KDP account did have the correct (and only) offset, but was a mobi7 not a kf8.
So the question of whether KDP corrects the error of having two text start offsets in the kf8 section of the uploaded file is still unanswered.