In preparation for moving JS to the shared data flow library, this PR removes all uses of barrier edges, replacing them with barrier nodes.

Theoretically it is not always possible to replace arbitrary barrier edges with barrier nodes, but for our use cases it seems to be possible. The shared library does not support barrier edges, and given that they can in practice become barrier nodes, there isn't a very strong use case for adding them.

The PR also adds isBarrierIn and isBarrierOut to the DataFlow::Configuration class, mimicking those from the shared library. These sometimes come in handy when replacing barrier edges.

Clear-text logging was a bit problematic as it opts out of the default PropRead taint step x -> x.prop by adding such steps as barrier edges. Neither ends of such an edge could be mapped to a barrier node, because the predecessor can have other outgoing edges, and the successor can have other ingoing edges. The solution I went with was to block the predecessor, but exclude cases where it might have other outgoing edges (and in those cases, we retain flow through the default x -> x.prop edge). It's not an elegant solution, though I'm not sure barrier edges were the best way to opt out of such taint steps in the first place. When and if we need a better solution, I'd rather work it out after transitioning to the shared data flow library.

Evaluation looks good.

• Some duplicate results from js/insecure-randomness have been removed/fixed. The new out-barrier more accurately handles what the barrier edge tried to do, which is to prevent flow out of sinks. The old barrier edge needed to restrict succ to avoid a blow-up, but this restriction only included local flow steps.