[DuplicateFinder] Consider project-specific similarity adjustments

[imnasnainaec]

Sometimes in a language, there may be a pair (or set) of letters such that substituting one for another commonly results in a different valid word. If there is such a pair of letters that are also not likely to be switched as the result of a typo (e.g., not near each other on the language's keyboard), then entries with vernacular forms that only differ by such a pair are not likely to be duplicates. Perhaps we could add a project setting for specifying pairs that should have altered weight in the duplicate finder algorithm. Word pairs with those differences wouldn't show up as early among sets of potential duplicates.

[imnasnainaec]

Questions:

• Can the Levenshtein distance be easily modified for this?
• While we're altering our distance algorithm, do we consider Damerau–Levenshtein distance (i.e., also allow transpositions)?