Unroll djbhash for ~25% speedup

This makes it about 25% faster and causes gcc's optimizer to make its own assumptions about length for the partial chunk at the end for better codegen too, believe it or not.