Scale roadmap graph traversal

[Deicyde]

Summary

• Cache containment and runtime node lookups while preserving the public graph models.
• Replace recursive and quadratic graph walks with iterative indexed traversal.
• Detect article mutation between graph loading and runtime projection.
• Preserve ordering, diagnostics, and legacy Graph pickle compatibility.

Validation

• 557 passed, 1 skipped
• Python 3.10 graph compatibility: 20 passed
• Randomized equivalence against the previous algorithms
• 25,000-node traversal stress test
• make lint
• make check-example

Reference implementation: #8

[Deicyde]

No blocking correctness findings at ceca8e8.

The new indexed traversal matched the parent algorithms on 20,000 randomized graphs, and 25,000-node dependency/containment probes passed across the supported Python matrix.

Non-blocking follow-up: bulk publication still constructs each container view separately and rescans all relations in graph_pages.py:98,134-138 and graph_views.py:195,214-236. A synthetic 2,000-container graph took 5.28 seconds versus 1.42 seconds at 1,000. This appears pre-existing, so it should not block this PR.