That way the specific configs which inherit from `AtmConfig` also inherit from `TaintTracking::Configuration`.
This removes the need for two separate config classes for each query.
All remaining functionality in `StandardEndpointFilters` is only being used in `EndpointCharacteristics`, so it can be moved there as a small set of helper predicates.
All remaining functionality in `CoreKnowledge` is only being used in `EndpointCharacteristics`, so it can be moved there as a small set of helper predicates.
`isOtherModeledArgument` and `isArgumentToBuiltinFunction` contained the old logic for selecting negative endpoints for training.
These can now be deleted, and replaced by a single base class that collects all EndpointCharacteristics that are currently used to indicate negative training samples: `OtherModeledArgumentCharacteristic`.
This in turn lets us delete code from `StandardEndpointFilters` that effectively said that endpoints that are high-confidence non-sinks shouldn't be scored at inference time, either.
This is needed because we changed the names of three endpoint filters that were all called "not a direct argument to a likely external library call or a heuristic sink" in order to disambiguate them (fc56c5a022).
Also disambiguate three filters from three different sink types that all have the same name, "not a direct argument to a likely external library call or a heuristic sink".
Implement the new query that selects data for training. For now we include clauses that implement logic that is identical to the old queries.
Include a temporary wrapper query that converts the resulting data into the format expected by the endpoint pipeline.
Move the small pieces of `ExtractEndpointData` that are still needed into `ExtractEndpointDataTraining.qll`.
