That's probably because the theorem assumes rational voters. For rational voters range voting colapses into tolerance voting for which Arrow's theorem IIRC does apply.
No it doesn't, not even if you do approval voting straight. (Which is a kind of range voting, just with a very small range of either 0 or 1.) To prove it for approval voting:
1. Independence of irrelevant alternatives: If N people approve of candidate X and M people approve of candidate Y and P people approve of candidate Z, then whether or not Z is on the ballot, X will still get a score of N and Y will get a score of M so X will still beat Y.
2. Montonicity: If a candidate changes his mind about whether or not he wants to approve candidate X, then X's total approval score will increase and everyone else's will remain constant, therefore X will not be hurt by voting for them.
3. Non-dictatorship: Obviously, since every voter contributes to the result equally, no person is dictator.
4. Unrestricted domain: It takes in a every single person's preferences and then deterministically produces a ranking of candidates by score.
5. Non-imposition: Any ranking of candidates is possible since if you want to rank candidates in the order X1 X2 X3.. all you have to do is have N voters approve of X1, N-1 approve of X2, and so on.
Therefore, approval voting satisfies the Arrow criteria. I imagine the general case of range voting can be proven in a similar way. Arrow's Theorem only describes deterministic functions which map a collection of individual preferences onto a single social preference, but since approval and range voting don't work quite that way it's okay.
(Also, it's not my understanding that Arrow's theorem assumes rational voting in the sense of trying to game the system by tactically voting. I think that even if voters can be expected to vote in a completely honest way Arrow's impossibility theorem applies to preference based voting systems.)
ETA: Technically, I think approval voting might violate unrestricted domain, but I think it seems to violate it which doesn't imply any injustice but that merely violates the mathematical assumptions underlying the theorem.