Graph Memory: canonical entity/fact temporal schema in Postgres

[aurexav]

Summary

Define the canonical graph memory model for ELF on top of Postgres as source of truth.

Problem

ELF currently stores notes and structured fields, but has no first-class entity/fact temporal model for graph memory.

In Scope

• Add additive Postgres schema for:
  • entities
  • facts (subject, predicate, object)
  • fact-evidence bindings
  • temporal validity (valid_from, valid_to)
• Define graph invariants and conflict semantics.
• Keep graph state auditable and evidence-linked.

Out of Scope

• No Neo4j or other graph database adoption.
• No multi-hop traversal/ranking algorithms.

Deliverables

• SQL migrations.
• Service-layer model and repository support.
• Docs with invariants and examples.

Acceptance Criteria

• Migrations are backward-compatible.
• Facts can represent active and superseded knowledge.
• Every fact is linked to at least one evidence-bearing source.
• Invariants are documented and testable.

Dependencies

• None.

Implementation Checklist

• Schema design review approved.
• Migration files merged.
• Read/write repo layer added.
• Invariant docs added.
• Integration tests for temporal validity and evidence linkage.

Done When

• A single canonical graph data model exists in Postgres and is ready for ingestion integration.

[aurexav]

Implemented in PR #64.

• Adds Postgres graph schema (entities, aliases, facts, evidence + validity).
• Wires tables into schema bootstrap.
• Adds storage helpers + ignored integration tests.

Verification:

• cargo make fmt-rust-check
• cargo make lint
• cargo make test-rust
• cargo make test-integration
• cargo make e2e