Add EdgesBetween and viz.FromQuery, support IN filters, revise docs and CI

[justintout]

Summary

• Query API:
  • IN and NOT IN filters now work and take a slice. They used to compile to x IN ?, which SQLite rejects. The slice is sent as one JSON array read with json_each, so the SQL text stays the same whatever the list length.
  • Return("labels") and Query.WithLabels() fill in Node.Labels. Labels stay off by default: returning them by default made traversals 30–50% slower.
  • EdgesBetween(ctx, nodeIDs, types...), on both Graph and Tx, returns the edges between a set of nodes. A unary + on target_id makes SQLite seek by source; without it, the plan did one index seek per source and target pair.
• viz:
  • FromQuery(ctx, db, q, options...) replaces NewFromResult. It runs the query with labels and draws every edge between the result's nodes.
  • Chart nodes now carry their graph ID. Echarts falls back to node names, so two nodes with the same name used to merge into one.
• Docs:
  • README updated for accuracy (dependencies, hop cap, pool options, update and transaction behavior, what results contain) and edited for style.
  • Benchmark results moved to BENCHMARKS.md.
• CI and tooling:
  • The test workflow now checks gofmt, vets the bench/compare module, and uses current action versions.
  • bench/compare uses a root go.work instead of a replace directive.
  • Added a // Package graph doc comment.

Testing

• Tests:
  • go vet ./... ./bench/compare/... and go test -race ./... pass.
  • New: TestWhereIn, TestEdgesBetween, TestFromQuery, TestRenderDuplicateNames.
  • TestReturn now covers labels.
• Query plans: checked with EXPLAIN QUERY PLAN for the labels subquery and EdgesBetween.
• Benchmarks: labels-by-default was compared against the base commit with benchstat, 6 runs each. That comparison is why labels are opt-in. Default queries compile to the same SQL as before.

🤖 Generated with Claude Code

https://claude.ai/code/session_01YB9owLYxU6NsNuVk2NVc7C