Ensure mean rate of setting bits near 0.5

[AshlinHarris]

https://anonlink-client.readthedocs.io/en/latest/tutorial/Blocking%20with%20Anonlink%20Entity%20Service.html#Inspect-the-output

[AshlinHarris]

#45 should contain a test comparing the diff of hash files on similar inputs

[AshlinHarris]

Test added in b340624, updated in 1790a7e

Regardless of schema settings, I couldn't get a value much lower than 0.7. I don't think we're setting too many bits, but I should go over the analysis linked above.

I wonder if changing half the bits means pseudorandomly reassigning half the bits, meaning 0.75 similarity. Would that still apply to further changes in the secret? The value was relatively consistent regardless of how different the secrets are.

[AshlinHarris]

What metric should be used to calculate bitstring distance? I haven't come across any relevant literature, but wouldn't maximum distance be about as big an issue as minimum distance?

Considering the theoretical worst case, if changing the input in some way leads to the new hash having a maximum Hamming distance from the old one, this would be a bad thing for cryptography. It means the two strings are just mirrors of each other, so one can be derived from the other by flipping all bits. It would also mean that two changes cancel each other out.

The actual ideal case is if a change can flip half of all bytes. I'm ignoring the distribution of the flips, which might be important but is out of scope for our current test plan.

[AshlinHarris]

The test added in db86a09 should be sufficient for comparing hashes from similar secrets. I think the docs linked at the top are focused on a different issue (number of bits set).