parallelized scrubbing

[deanmalmgren]

Right now, we scrub for each feature (email, credentials, URLs, names, etc) in a particular order. This order has been fine tuned to address certain things that have come up during development, but its easy to imagine this dependency tree becoming ever more complicated and ultimately becoming very difficult to tune for all possible examples.

A different approach would be to do all of these scrubbings in parallel instead of in series. For example, we could have a method like Scrubber.iter_filth() that yields a Filth object (based on a MatchObject?). The iter_filth method could be monitoring all of the features we are scrubbing and return them in the order in which they appear in the text while resolving any conflicts that come up along the way. Each Filth object, for example, could have a Filth.score attribute (or property or method) that signals the likelihood that this piece of Filth is actually a piece of CredentialFilth and not EmailFilth.

This would allow us to properly deal with situations like this:

Your credentials are the following
username: joe@example.com
password: p@ssw0rd

which (I think) would preferrably yield

Your credentials are the following
username: {{USERNAME}}
password: {{PASSWORD}}

or optionally

Your credentials are the following
username: {{EMAIL}}
password: {{PASSWORD}}

but certainly NOT

Your credentials are the following
username: {{EMAIL}}
password: p@ssw0rd