Deliver reproducible FoodGraph knowledge-graph demo

[teopaleface]

Outcome

Delivers the complete FoodGraph vertical slice for the one-day team project:

• frozen 130-product Open Food Facts fixture and deterministic validation;
• local-only Neo4j 2026.06 schema, provenance graph, idempotent import and guarded reset;
• 130 committed 768-dimensional nomic-embed-text vectors;
• graph-expanded grounded retrieval and fixture-bound offline fallback;
• displayed, allowlisted, read-only Text2Cypher with EXPLAIN and reviewed fallback;
• auditable Product → Fact → Source/Agent traversal;
• clean Jupyter demo using the environment kernel;
• CI, smoke checks, four-person roadmaps and release handoff.

Verification

• python scripts/validate_fixture.py — 130 products, 7,490 facts, fixture SHA e383aff…
• python scripts/validate_artifacts.py — 130 × 768 vectors, exact fixture/model binding
• python -m unittest discover -s tests -v — 52 passed
• ruff check and ruff format --check — passed
• graph import run twice — identical counts, vector index ONLINE
• embeddings import — 130 products
• smoke test — passed
• live and forced-offline notebook execution — passed with 0 error outputs
• tracked-file secret scan and local-link/config validation — passed

Independent implementation review found no P0/P1 blocker. The initial integrated PR is the bootstrap exception recorded in D-008. FG-013 and FG-014 remain open for the physical second-laptop rehearsal and final tag.

Closes #1
Closes #2
Closes #3
Closes #4
Closes #5
Closes #6
Closes #7
Closes #8
Closes #9
Closes #10
Closes #11
Closes #12