ddg edges reference endpoints that are not body nodes of their own callable

[rahlk]

Problem

Some ddg edges name an endpoint key that does not exist in the callable's own body{} map. On daytrader8 at -a 4, 87 of 5,434 ddg edges dangle this way, across 38 distinct keys.

Every one of them shares a shape:

• the key is always <line>:0 — column zero, which no real body-node key uses
• every affected edge carries prov: ["points-to"]
• cfg (6,984 edges) and cdg (4,416) have no such endpoint, and neither direction of ddg dangles for any other provenance

So it looks specific to the points-to pass minting a key that the body map never received, rather than a general keying problem.

Consequence

The Neo4j emitter materialises body nodes from body{}, so an edge whose endpoint was never a body node cannot be written: the graph holds 5,347 where the JSON claims 5,434. That is the only place a consumer reading both artefacts sees them disagree — everything else round-trips exactly.

Found while verifying that an SDK's two backends answer identically. Rather than tolerate a fuzzy count, the test now asserts the difference is exactly the set of dangling edges, so a change in this behaviour surfaces immediately.

Repro

java -jar codeanalyzer-3.0.1.jar -i <daytrader8> -a 4 -o out

then for each callable, check every ddg edge's src and dst against the keys of that callable's own body{}. The 87 misses are all <line>:0.

Not visible on a small pruned project — a fixture of four files has none — so a scale corpus is needed to see it.

Definition of done

• Every ddg endpoint is a key of the callable's own body{}, or the vertex it names is emitted.