Canonical Neo4j projection contract: merge labels, body-node model, can:// grammar

[rahlk]

Parent: #35 · Roadmap: docs/design/roadmap.md (candidates 1, 2, 4)

Skeleton — bodies to be filled in by the maintainer.

Problem

Canonical v2 is specified, but each analyzer projects it differently and nothing states which projection is correct. codeanalyzer-python uses 9 Neo4j merge labels and splits the body model across PyCallSite and PyCFGNode; codeanalyzer-typescript uses 2 (Application, CanNode) with a unified TSBodyNode; codeanalyzer-java has no statement-level nodes in Neo4j at all (.claude/SCHEMA_DECISIONS.md #9) and no can:// ids.

Scope boundary

Writes the contract. Does not migrate any analyzer, does not build the conformance suite or the parity gate (separate sub-issues), and does not settle the entrypoint vocabulary (its own design session).

Goals

• Neo4j projection contract: merge-label strategy, language namespacing, what "always full-depth" obliges
• Unified body-node model — one node with a kind discriminant, and what Java must add
• can:// grammar conformance: per-language signatureOf() pinned, including Java which has none today
• Committed to docs/design/specs/

Caveats and known risks

• TODO
• Java's statement-level CPG was deliberately deferred, not overlooked — the contract must state what it now requires and at what cost.

Definition of done

• TODO — exact conditions, not "spec written".