Add opt-in BreadthFirst traversal mode

[justintout]

Stacked on #2.

Summary

Adds Query.BreadthFirst(), which makes the most recent Related step expand each reached node once. Results are identical to the default mode.

g.Match("Person").
    Where("name", "=", "Alice").
    Related("KNOWS", 1, 6).
    BreadthFirst().
    Run(ctx)

Why opt-in

By default, a multi-hop step is a recursive CTE that deduplicates on (node, depth). A node reachable at several depths gets expanded once per depth. BreadthFirst compiles the step to one CTE per hop level, holding only nodes first reached at that level.

Each level costs a fixed ~20µs for SQLite's temporary tables. Which mode wins depends on how often the walk revisits nodes. That's a property of the data near the start node, and the query shape can't predict it. Choosing automatically would mean running a probe query on every deep traversal.

Medians of 6 runs, Apple M3 Max:

Query	Default	BreadthFirst
100-node chain, 10 hops	40 µs	243 µs
Tree, fanout 3, depth 6	1.81 ms	1.67 ms
Dense 500 nodes / 5 edges each, 5 hops	2.10 ms	1.44 ms
Dense 500 nodes / 5 edges each, 10 hops	9.51 ms	2.02 ms
10k random, 6 hops	1.87 ms	1.68 ms
10k random, both directions, 4 hops	2.28 ms	1.78 ms

Details

• minHops: requires minHops == 1; otherwise Run returns an error. A larger minHops needs nodes reachable at an exact depth, and a visited set discards those.
• maxHops == 1: compiles as a normal single hop.
• Cycles: start nodes aren't excluded from later levels, so a cycle back to the start is found, matching the default mode.

Test plan

• go vet ./... and go test ./... pass.
• TestTraversalMatchesReference runs every minHops == 1 case in both modes, for all directions and 1–6 hops, against walks computed in Go on a random graph with cycles and self-loops.
• Breaking the last breadth-first level on purpose makes that test fail.
• TestWhereRelIntermediateStep covers BreadthFirst followed by WhereRel and another step.
• BreadthFirst with minHops of 2 returns an error.

🤖 Generated with Claude Code

https://claude.ai/code/session_01YB9owLYxU6NsNuVk2NVc7C