feat(java): view dispatches — JViewDispatch models, get_view_dispatches / get_view_dispatchers, Neo4j reconstruct; pin codeanalyzer-java 3.3.0 #404

Plan (optional)

Spec: codellm-devkit/.github docs/design/specs/2026-09-11-java-view-templates-and-dispatch.md § 7 (PR codellm-devkit/.github#80). Analyzer side: codellm-devkit/codeanalyzer-java#259, released as codeanalyzer-java 3.3.0.

Problem

codeanalyzer-java 3.3.0 emits application.view_dispatches[] {src, dst, via, prov} and view_dispatches_unresolved[] {site, callee, target, via, reason, prov}, and projects J_DISPATCHES_TO {via, prov} from :JBodyNode to :Artifact. The SDK's JApplication (cldk/models/java/models.py) declares neither list — they parse away silently under extra="ignore" (#386) — the Java facade has no accessor for "which pages can this code reach", and the Neo4j reconstructor does not read the relationship back. The SDK also still pins codeanalyzer-java==3.1.2 (pyproject.toml:46,89), two releases behind; 3.2.0 added module.source, span byte offsets, body_span, type parameters and call-site facts that the pin has not picked up either.

Scope boundary

The Java view-dispatch surface and the pin bump only. No template-internal nodes (roadmap Group F), no routes (Group E), no change to python/typescript facades — their analyzers emit no dispatches yet. No new backend ABC beyond the two accessors; unresolved dispatches stay JSON-only, as config_reads_unresolved is.

Goals

• Models: JViewDispatch {src, dst, via, prov} and JViewDispatchUnresolved {site, callee, target: Optional[str], via, reason, prov}; JApplication.view_dispatches: Optional[List[JViewDispatch]] = None and .view_dispatches_unresolved, mirroring config_uses / config_reads_unresolved.
• Facade: JavaAnalysis.get_view_dispatches(view: str | None = None) -> List[JViewDispatch] (filter by artifact path suffix when view is given) and get_view_dispatchers(path: str) -> List[JCallableOverview] (the callables owning the dispatching body nodes), beside get_config_uses / get_config_readers; backend ABC + local + Neo4j implementations.
• Neo4j reconstruct: read J_DISPATCHES_TO (via, prov) back into view_dispatches; document that view_dispatches_unresolved is JSON-only.
• Pin: codeanalyzer-java==3.3.0 in dependencies and [tool.backend-versions].
• Docs: the Java facade reference gains the two accessors.

Caveats and known risks

• callee on an unresolved record is an @external id for a call site but the enclosing callable's id for a return site (spec § 5) — model it as a plain string, not externalCanId.
• via is a closed set today (forward | include | redirect | view-name | navigation) but navigation is reserved and never emitted by 3.3.0; a Literal would be correct now and would break on the JSF work later — prefer str with the vocabulary in the docstring, as JConfigRead.reason does.
• The 3.1.2 → 3.3.0 pin jump also lands 3.2.0's additive fields; the e2e test fixture output changes accordingly.

Definition of done

• The mocked test parses an analysis.json carrying both lists and round-trips them unchanged; get_view_dispatches("home.jsp") returns exactly the hand-listed edges and get_view_dispatchers("src/main/webapp/home.jsp") exactly the owning callables.
• Backend-contract test: local and Neo4j backends agree on get_view_dispatches() over the same run.
• E2E against the 3.3.0 binary: the DayTrader-style servlet fixture yields J_DISPATCHES_TO rows equal in number to view_dispatches.
• pytest tests/test_java_analysis.py tests/test_java_backend_contract.py tests/test_java_e2e.py green; pins moved in both places.