Instead of modeling individual implementations, take a more general
approach of reuse dataflows for interfaces defined by a library. This allows
tracking flows across all implementations and aligns better with how we
manually model frameworks. This may have some FPs given all possible flows
are modeled for a specific interface but also covers more scenarios where
we don't know which implementation of an interface is used.
Avoid generating models for types that can't really propagate taint
in a valuable way (e.g. primitivies, BigInt, ..). Keep tracking
bulk-like data (e.g. char[] or byte[]).
