Asymptotically speed up range checker chip trace generation

[morganthomas]

The implementation of Chip<M>::generate_trace for RangeCheckerChip<MAX>, in range/src/lib.rs, is O(MAX * log(|self.count|)) but can be done in O(|self.count| * log(|self.count|)). In this context, self.count is a BTreeMap<u32, u32> whose key space is 0..MAX. This means that |self.count|, the cardinality or number of keys of self.count, is less than or equal to MAX, and thus the algorithm is asymptotically slow, as noted by the code comment.