Neo4j projection omits import/export bindings, callable parameters, and config-read edges

[rahlk]

Problem

Three gaps in the Neo4j projection, each of which makes a python-sdk accessor refuse rather than answer. All three are recoverable from data the analyzer already computes for analysis.json. Measured on main @ 2064b6b (cants on itself: 139 modules, 1008 callables):

gap	analysis.json	graph
1. import / export bindings	801 imports, 67 exports	none — get_imports() / get_exports() refuse
2. callable parameters	1546 parameters on 970 callables	none — get_method_parameters() refuses over Neo4j, answers in process
3. unresolved config reads	config_reads 10 (artifacts-app, -a 2)	none — TS_USES_CONFIG (resolved reads, :TSBodyNode → :ConfigKey) IS projected; only the unresolved counterpart is missing

Row 3 corrects this issue as first filed: TypeScript does not emit DEFINES_CONFIG alone.

language	config relationship types
Python	DEFINES_CONFIG, PY_USES_CONFIG, PY_READS_CONFIG_UNRESOLVED
TypeScript	DEFINES_CONFIG, TS_USES_CONFIG
Java	DEFINES_CONFIG (codeanalyzer-java#231)

Design

Spec: docs/design/specs/neo4j-bindings-parameters-config.md (decisions D1–D6, also in .claude/SCHEMA_DECISIONS.md). Additive on Neo4j contract 2.0.0; ships in 1.4.0.

Scope boundary

In scope: the three rows above. Out of scope: #179 (:TSCallable.code one line short), #180 (structuredClone ceiling), #181 (exit code on failure), #177 (declaration merging) — all filed separately.

Goals

• D1 resolved_module on TSImport / re-export TSExport (ts-morph resolution at build time)
• D2 TS_IMPORTS aggregated per (importer, target): spellings[], imported_names[], aliases[], type_only_names[]; externals to the @external/<root> ghost
• D3 exports_json on :TSModule + TS_RE_EXPORTS aggregated like D2
• D4 parameters_json on :TSCallable, python encoding, null when empty
• D5 TS_READS_CONFIG_UNRESOLVED :TSApplication → :TSExternal | :TSCallable, _k = key|reason
• schema.ts + regenerated schema.neo4j.json; conformance, edge-identity and bolt tests
• consumer skill docs/skills/analyzing-cants-graphs/ documents the three query shapes

Caveats and known risks

• Parity clause: python fixes the vocabulary for D2/D4/D5; D3 is coined here (no sibling has ES-module exports). Java #231 must spell JAVA_READS_CONFIG_UNRESOLVED the same way.
• Size cost (measured): full python-shape parameters_json = 430 KB on cants self = 1.4 % of graph.cypher. Accepted over a minimal shape (148 KB) to avoid a second parameter vocabulary.
• TS_READS_CONFIG_UNRESOLVED inherits python's per-site collapse (_k = key|reason, no site): counts differ from analysis.json, presence/absence agrees.
• Declaration merging mints one can:// id for a value and a type of the same name; the Neo4j MERGE collapses them into one node #177: parameters_json is a property on the existing node, so a merged value/type node keeps the last writer's list exactly as it keeps the last writer's kind — not worsened, not fixed.

Definition of done

• The three python-sdk accessors that refuse on a TypeScript graph answer, with no SDK change beyond reading the new data (the SDK's reconstruct.callable_ already decodes parameters_json), and the SDK's documented tolerances for them can be deleted.
• Snapshot on test/fixtures/sample-app shows TS_IMPORTS to :TSModule for the four relative specifiers and to ghosts for commander, neo4j-driver, node:*; artifacts-app shows one TS_READS_CONFIG_UNRESOLVED per distinct (callee, key, reason) in config_reads.
• L1 ⊆ L2 ⊆ L3 ⊆ L4 gate green; bun test green; schema.neo4j.json regenerated.