This patches out bytecount's "fast" vectorized algorithm on big-endian
machines, where it has been observed to fail. Going forward, bytecount
should probably fix this on their end, but for now, we take a small
performance hit on big-endian machines.
Fixes#1144