* Instead of keeping md5 samples separate, merge them as one hash computed from the various selected chunks we picked.
* We don't need to keep a boolean to see whether or not the user chose to optimize; we can simply compare the value of the threshold, since 0 means no optimization currently active.