Batch neighborhood traversal queries

[jithinmathws]

Summary

• batch neighborhood inputs and adjacency loading by BFS traversal level
• reduce depth-two neighborhood database queries from 150 to 12
• preserve sequential BFS admission, deterministic ordering, scoring, explanations, uncertainty handling, graph closure, and response schema
• add batch-equivalence and bounded-query regression tests
• record remediation benchmark evidence and acceptance results

Performance validation

Production-count synthetic fixture:

• 1,727 materials
• 4,144 material-element associations
• 128 material-application associations
• request: material_id=1, depth=2, limit=25

Results:

• database queries: 150 → 12
• warm median service latency: 1,934.218 ms → 300.305 ms
• 100 ms simulated query RTT: 17,133.475 ms → 1,543.270 ms
• serialization: 0.244 ms
• payload: 9,657 bytes, unchanged
• payload SHA-256: 50f22e6a428bd19f7569f169361e71cf5d019056c23e07d75cf2af97a2f2fd11, unchanged

Validation

• focused neighborhood tests: 15 passed
• complete suite: 936 passed, 1 skipped
• Ruff passed
• dependency contract passed
• immutable automation reference validation passed
• git diff --check passed
• repeated benchmark responses were deterministic

Boundaries

No production access, deployment, restart, schema change, migration, index change, provider change, or concurrency claim is included.