test(rel): prove expression and relational-lowering boundaries

[DecisionNerd]

Problem\n\nThe accepted #360 workspace report shows the largest concentrated executable gaps in graphforge-rel/src/expr.rs and graphforge-rel/src/lowerer.rs. Existing happy-path and TCK coverage does not directly prove malformed Arrow shapes, heterogeneous value access, temporal boundaries, or created-row schema rejection behavior.\n\n## Objective\n\nAdd deterministic table-driven and property tests for reachable relational expression and lowering contracts without changing product behavior.\n\n## Requirements\n\n- Cover success, exact failure, and meaningful boundary cases for dynamic list/map/property access, heterogeneous values, temporal conversion/order, and scalar normalization.\n- Cover created-row schema, reserved topology fields, reference-vs-minted entities, and malformed/unbound lowering inputs.\n- Prefer public lowering behavior where it proves multiple layers; use module tests for pure Arrow/DataFusion helpers.\n- Do not exclude code, manufacture product branches, or accept missing results.\n\n## Acceptance Criteria\n\n- [ ] Selected uncovered expression/lowering contracts have explicit oracles.\n- [ ] Representative wrong value, wrong error, null, empty, overflow, and malformed-schema mutations fail.\n- [ ] graphforge-rel remains above the per-crate floor and materially reduces the canonical aggregate deficit.\n- [ ] Targeted and workspace gates pass with no runtime behavior change.\n\n## BDD Completion Scenarios\n\n### Scenario: Dynamic access rejects malformed values exactly\n\nGiven null, valid, wrong-type, out-of-range, and malformed tagged Arrow values\nWhen relational expression evaluation runs\nThen valid inputs return exact values\nAnd invalid inputs return the documented DataFusion error class.\n\n### Scenario: Created rows have a deterministic schema\n\nGiven reference and newly minted node specifications\nWhen relational lowering builds the emitted-row schema\nThen input columns and minted identity/property columns have exact types and qualifiers\nAnd reserved or unresolved fields are rejected.\n\n## Testing\n\nUse table-driven unit/property tests plus real lowering integration tests. Run the canonical coverage ledger and standard Rust gates.\n\n## Documentation\n\nUpdate testing evidence mapping only if the canonical coverage documentation needs a new boundary example.\n\n## Non-Goals\n\n- Executor persistence behavior.\n- Storage publication and facade contracts.\n- Runtime feature changes.\n\n## Related Issues\n\n- Parent and canonical close gate: #360\n- Blocked by the foundational coverage-ledger child.\n

[coderabbitai]

🔗 Related PRs

#220 - feat(bindings): expose embedded write modes [merged]
#247 - feat(iac): validate static deployment targets [merged]
#351 - fix(bindings): repair Node bulk API contract [merged]
#358 - test(api): make public BDD fail closed [merged]
#361 - test(ci): measure native binding Rust coverage [open]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!