@ruvector/graph-node: storagePath data silently lost on restart (open() never hydrates, edges never persisted)

[Bolivar1234]

Summary

@ruvector/graph-node (v2.0.4) accepts a storagePath and writes node records to redb, but the NAPI binding never wires the rest of persistence, so any graph silently loses edges — and on reopen, everything — after a process restart:

1. GraphDatabase.open(path) never hydrates. It opens the redb file but returns an empty graph (stats() → {totalNodes: 0, totalEdges: 0}) even when the file contains data. The crate already has GraphDB::with_storage() / load_from_storage(); the binding just never calls them.
2. createEdge / createHyperedge / batchInsert edges never reach storage. Only nodes are persisted; every edge lives solely in the in-memory hypergraph.
3. deleteNode doesn't cascade through storage, leaving dangling persisted edges.

Repro

const { GraphDatabase } = require('@ruvector/graph-node');
const Z = new Float32Array(4);

// process 1
const db = new GraphDatabase({ distanceMetric: 'Cosine', dimensions: 4, storagePath: './t.db' });
await db.createNode({ id: 'a', labels: ['T'], properties: {}, embedding: Z });
await db.createNode({ id: 'b', labels: ['T'], properties: {}, embedding: Z });
await db.createEdge({ from: 'a', to: 'b', description: 'CALLS', confidence: 1, embedding: Z });
console.log(await db.stats()); // { totalNodes: 2, totalEdges: 1 }

// process 2 (restart)
const db2 = GraphDatabase.open('./t.db');
console.log(await db2.stats()); // { totalNodes: 0, totalEdges: 0 }  <-- data gone

Root cause

In crates/ruvector-graph-node/src/lib.rs:

• new()/open() construct GraphDB::new() and never replay storage into the hypergraph or property graph.
• create_edge/create_hyperedge/batch_insert only call hypergraph.add_hyperedge(...); there is no storage.insert_edge/insert_hyperedge write-through.
• delete_node deletes only the node record from storage.

The storage layer itself (ruvector-graph's redb backend, all_node_ids/all_edge_ids/all_hyperedge_ids, insert_edge, insert_hyperedge) works correctly — it's purely a binding gap.

Fix

I have a working patch (PR to follow, referencing this issue) that:

• hydrates hypergraph + property graph in new()/open() (nodes, binary edges, multi-node hyperedges; embeddings/confidence round-tripped as __embedding/__confidence FloatArray properties)
• writes edges/hyperedges through to storage in createEdge/createHyperedge/batchInsert
• makes deleteNode cascade through persisted edges/hyperedges and deleteEdge clean the traversal index
• makes hydration tolerant of dangling records (skip + report) rather than failing open()

Verified with a 12-gate suite (create / reopen in fresh process / delete persistence / batch / tx / Cypher / kHop / in-memory mode / single-writer lock) and a real workload: a 154,711-node / 353,496-edge code graph that now survives restart with exact counts (hydrate ~15s).

Happy to adjust the approach if you'd prefer hydration to live in the crate rather than the binding.