Add vector fingerprint visualization to point cards

[generall]

What

Points in the collection view now show a compact "fingerprint" strip next to each vector's Length chip — a 32-bar waveform where similar vectors produce visually similar silhouettes. Covers dense vectors, multivectors and sparse vectors, in both the Points tab and the graph-view point preview (shared PointVectors component).

How

• src/lib/vector-fingerprint.js — each bar is a SimHash-style projection of the whole vector onto a deterministic random ±1 direction (one murmur-finalizer hash per dimension; its 32 bits are the projection signs, so it stays a single O(dim) pass even for 3000+-dim vectors). Values are normalized by the vector's L2 norm, making the fingerprint magnitude-invariant.
• Sign stability guarantee from SimHash: for cosine similarity s, a fraction 1 − acos(s)/π of bars keep the same sign — ~86% at cosine 0.9, ~50% for unrelated vectors.
• Multivector rows get row-specific sign patterns (no cross-row cancellation); sparse indices are hashed into the same scheme.
• src/components/Points/VectorFingerprint.jsx — renders a mirrored waveform on a small DPR-aware canvas: bars up in Qdrant red for positive projections, down in blue for negative, around a faint center line. Theme-aware (lighter shades in dark mode). Unsupported shapes (e.g. {text: ...} inference objects) render nothing.

Testing

• 8 unit tests: determinism, scale invariance, values in [-1, 1], closer-vectors-give-closer-fingerprints, >80% sign agreement under small perturbation, sparse/multivector/small-dim shapes, unsupported-shape guards.
• eslint, prettier and npm run build all pass.

🤖 Generated with Claude Code