This is not supported by any of the sources in the wikipedia page. There don't seem to be any sources saying how the sample was collected from each or which counties. ...you are right it does not say "select".
But regardless, a margin of ~100 votes on a sample of 175K votes is not a margin with which you can make the statement "it seems pretty safe to say that a 100% accurate tally of the votes in Florida would have given the state to Gore" ...that's ludicrous.
We know that the butterfly ballots were a poor design that caused vote confusion, and we also know that Bush was less affected by this than Gore (by virtue of the fact that Bush was the first hole, and thus any misalignment caused by an oblique view would line up with no hole, as opposed to oblique views causing Gore's arrow to line up with Buchanan's hole).
From the actual results, we know there is a chunk of votes in Palm Beach for Buchanan that doesn't seem appropriate. According to Buchanan's campaign, this would be a bit shy of 3000 votes for Buchanan that should have been for Gore [1]. No recount would have changed these votes (nor should one), but a more accurate voting machine would have given these votes to Gore.
That's why I say that it seems safe to say that Gore legitimately won Florida, on the basis on what the voters intended to vote, but it's quite unclear what even the best recount would have said the winner is.
[1] https://en.wikipedia.org/wiki/Pat_Buchanan#2000_presidential...