Tolerance rate in Complishon

[omarabedelkader]

This PR adds typo-tolerant completion to the completion engine.
At the moment, completion depends on exact prefix matching. This means that a small typo can prevent the expected completion from appearing, even when the intended candidate is obvious. That makes completion less helpful in one of the situations where it should help the most: while the user is actively typing and makes a minor mistake.

To improve this, this PR introduces approximate matching based on the normalized Damerau-Levenshtein distance. When enabled, the completion engine can tolerate small typing errors and still propose close candidates. Exact prefix matches are preserved and still ranked first, while fuzzy matches are only considered as a fallback and are sorted by similarity.

For example, CoCompletionTolerance rate:0.4

• if the intended varibale is Color and the user types a close but incorrect form such as Clor or Colr, completion can still suggest Color

or

• if the intended selector is printOn: and the user types a close but incorrect form such as prntOn:, completion can still suggest printOn:.

P.S: The main trade-off is that increasing the tolerance too much can cause incorrect matches, so completion becomes noisy.

[omarabedelkader]

By default the tolerance rate is 0.0 to be able to pass the tests

[omarabedelkader]

Failed tests are unrealated

[Ducasse]

I'm wondering if we could not add a strategy and delegate to it. Like that we could avoid the duplication of the call chain.