Document assumptions about the input error characteristics

[hardbyte]

As each PII field may have different types of errors (e.g. missing data, transcription error, entirely changed data...) we need to document any built in assumptions.

• Matching common names: "Wu", "Smith" etc
• Switching fields: "Dexter Cody" vs "Cody Dexter"
• Edits which preserve letter frequency but not bigrams: "Gerg" vs "Greg"
• Address changes: "123 blah lane" vs "84 another street"
• Format issues: "14/8/2018" vs "8/14/2018"

I've put the issue here but it might make more sense in clkhash.

cc: @wilko77 @nbgl

Aha! Link: https://csiro.aha.io/features/ANONLINK-13