codeanalyzer-java: L2 v2 call_graph (identity edges + external_symbols) #181

Plan (optional)

Design spec: https://github.com/codellm-devkit/codeanalyzer-java/blob/main/docs/design/specs/schema-v2-l3-l4-design.md

Problem

v1 emits rich JGraphEdges (embedding method detail) as the call graph. Canonical v2 requires an identity-only call_graph at application scope and the callee: null → id refinement on body call nodes.

Scope boundary

L2 edges + callee backfill only, over the L1 v2 tree. WALA RTA call-graph construction is unchanged (Java's only call graph is WALA's). No dataflow.

Goals

• Identity-only call_graph:[{src, dst, prov, weight}] at application scope; endpoints are callable can:// ids
• Backfill each body call node's callee (null → id) on resolved sites
• external_symbols{} homes call targets outside the project (imported/library/builtin), keyed by can://…/@external/…
• Edge-only-when-resolved: unresolved sites keep callee:null, skip the edge, never crash

Caveats and known risks

• WALA-RTA unsoundness (reflection, dynamic dispatch, unmodeled natives) — document in the analyzer README; precision recovered downstream by the SDK.
• call_graph edges are immutable once written (never re-anchored to a statement at L3).

Definition of done

• L2 gate: no dangling endpoints (every src/dst resolves to a real callable id); every edge has non-empty prov; callee backfilled on resolved sites and still null on honest-unresolved ones; a named expected (src,dst) edge present + at least one cross-package edge.
• L1 ⊆ L2 superset holds (nothing L1 emitted changed except the callee backfill); output validates against the SDK model.