python-sdk: Java models → schema v2 (two-layer views)

[sinha108]

Plan (optional)

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

Problem

The python-sdk Java models are v1 (JApplication{symbol_table, call_graph, system_dependency_graph}, rich JGraphEdges). They must load the analyzer's v2 output while keeping every public accessor's name, signature, and return type unchanged.

Scope boundary

The SDK Java model + view layer only — no analyzer changes, and no slice/taint/reachability queries (those are the separate SDK-queries issue). Starts after the analyzer's v2 release is cut (until then the old models can't parse v2). Reconcile the open cldk/models/cpg/ vs per-language projections.py question (seen on release/2.0) here — the shared layer likely already exists from the Python/TS v2 migration, so Java adds its views onto it.

Goals

• Load the v2 envelope + CPG; Java views preserving accessor names/signatures/return types (.code/.call_sites become computed views over body/source)
• AnalysisLevel gains L3/L4; new get_program_dependency_graph; get_system_dependency_graph returns the v2 SDG
• Document the sanctioned semantic shifts (nx node keys → can:// ids; rich edges retire; envelope keys change; call-site payload thins)
• Pin the analyzer v2 version once its release is cut

Caveats and known risks

• Major SDK version bump, but not a breaking public API — names/signatures/return types frozen behind the view layer.
• Consumers reaching past the public API (into nx node-key types or JSON string internals) can break — document each shift.

Definition of done

• SDK model gates green: mocked + E2E against the facade load v2 output; public API unchanged (or the major bump + shims documented).
• Parity with the committed spec; analyzer↔SDK versions pinned in lockstep.
• (Slice/taint query gates are the separate SDK-queries issue, not this one.)