fix(ruvector-cli): real demo + binding-drift fixes for v0.2.25 (#400, #402)

[ruvnet]

Follow-up to #403. Fixes the runtime-side issues from #400 (ruvector demo modes) and #402 §A/§B (VectorDB CRUD + GNN/attention typed-array errors) — the bits that needed binding-surface investigation.

Changes

VectorDBWrapper: normalize distance metric (§A root cause)

Native JsDistanceMetric enum is PascalCase (Cosine | Euclidean | DotProduct | Manhattan); CLI passes lowercase shorthand. Added normalizeMetric() in src/index.ts that maps casing + common aliases (l2, dot, dotproduct, innerproduct, l1) and accepts metric as a constructor alias for distanceMetric.

demo --basic: realign with current API (§A surface)

Old: db.insert('vec1', [1.0, 0.0, 0.0, 0.0], { label: ... }) (positional).
Now: await db.insert({ id, vector: new Float32Array(...), metadata: {...} }) — matching the wrapper's actual signature.

demo --gnn: typed-array fix + binding-bug surface

• Fixed: pass Float32Array everywhere; handle both result.weights (wrapper) and result.attention_weights (older shape).
• Surfaced: @ruvector/gnn@0.1.25 has a published-binding regression where every method throws Given napi value is not an array regardless of input shape. Added reportGnnBindingError() helper that detects the pattern and points at ruvector@0.2.23: data-plane broken — VectorDB CRUD, GNN, attention, embed, rvf, benchmark all fail #402.

demo --graph: real round-trip via GraphDatabase

Was "coming soon" stub. Implemented Alice -[:KNOWS]-> Bob using createNode({id, embedding, properties}) / createEdge({from, to, description, embedding, confidence}).

demo --benchmark: real inline benchmark

Was a redirect to ruvector benchmark. Implemented 1000-insert / 100-query mini-bench. Pinned to dim=4 because ruvector-core@0.1.29 has a regression that ignores the dimensions constructor arg — once the binding is rebuilt, dim can scale up.

attention compute: align with current API (§B)

CLI invoked attn.forward([query], keys, values). Current @ruvector/attention classes expose compute(query, keys, values) — forward doesn't exist. Replaced all five branches; query must be flat Float32Array, not [query]. Reproduces and fixes user's Undefined into rust type u32 / Get TypedArray info failed errors.

gnn layer --test / compress / search: typed-array conversion

All three previously passed number[] where the binding needs Float32Array. Converted at call sites + wired reportGnnBindingError so users see the upstream pointer when they hit the binding-side regression.

Verification

$ node bin/cli.js demo --basic
  Searching for nearest to [0.8, 0.6, 0, 0]:
    1. vec4 (score: 0.0101)  ✓ correct nearest
    2. vec1 (score: 0.2000)
    3. vec2 (score: 0.4000)
  Demo complete!

$ node bin/cli.js demo --graph
  ✓ Created nodes: Alice, Bob
  ✓ Created edge Alice -[:KNOWS]-> Bob
  Graph demo complete!

$ node bin/cli.js demo --benchmark
  ✓ Inserted 1000 vectors in 126ms (0.13ms/vec)
  ✓ 100× top-10 search in 51ms (0.51ms/query)

$ node bin/cli.js attention compute -q \"[1,0,0,0]\" -k keys.json -t dot
  ✔ Attention computed (dot)

Version bumped 0.2.24 → 0.2.25.

Out of scope (binding-side rebuilds tracked separately)

Two published native bindings have regressions that the CLI can detect but not fix:

Binding	Symptom	Action
@ruvector/gnn@0.1.25 (and platform pkgs)	Every method throws Given napi value is not an array	Needs rebuild from current source
ruvector-core@0.1.29 (and platform pkgs)	Constructor dimensions arg ignored, all inserts pinned to dim=4	Needs rebuild from current source

The Rust source in this repo shows correct behavior — the published .node artifacts are stale. Rebuild + republish is tracked at #402.

Closes #400 (demo modes runtime). Partially addresses #402 (§A/§B CLI-side fixed; binding-side rebuilds tracked).