This task covers the technical implementation of the Multilingual Revert Risk model in Automoderator. We should be able to choose whether this model or Language Agnostic is used per wiki (though this shouldn't be in Community Configuration). Note that this will impact liftwing client requests as well as ORES score lookups.

We will also need different score thresholds for our caution levels with this model (scores are pending T372747).

The model will be a variable that defaults to language-agnostic and can be changed. The language-specific scores will live in the extension.json file.