* Add test cases
* Change preprocessing for autoclassification
* Change patterns of recognizers
Affected recognizers:
- `CreditCardRecognizer`: also included a sanitation step before analyzing
- `UsLicenseRecognizer`: pattern that avoids version strings like `V1`
- `AuTfnNumber`: be more specific with patterns
* Remove normalization in processor
The goal is to remove a normalization that doesn't make much sense in an environment where we may get only one result.
Use `tolerance` to filter noise and depend on `confidence_threshold` to remove low score results
* Fix tests