[New1Direction]

Clean integration of PR #23

Integrate tested server-sliced graph delivery onto current main after source discovery/imports and the adaptive mathematical layout were merged. No new infrastructure or frontend rewrite.

Changes

• Local GET graph endpoints return one atomic compact status summary and searched/paginated view (<=90 nodes, <=300 internal edges, matching evidence only). Selected component evidence is separately capped at 40 adjacent edges.
• Existing adaptive local control room now requests graph pages and node detail on demand, preventing old search responses from overwriting current results.
• Fixed graph/incident summary explicitly distinguishes missing/stale/unknown health; no inferred runtime status.
• Python projection tests, real Chrome 405-node Compose test, and real read-only repo.ing payload benchmark.

Measured historical baseline

Previous real repo.ing smoke measured 659,045 bytes full graph vs 85,943 bytes atomic summary+first-page (~86.96% smaller initial payload, not render speed). Recheck against current main CI before merge.

Merge criteria

All Python 3.11/3.13, mathematical solver/Chrome and actual repo.ing smoke green. Keep loopback only; no hosted tenant auth, automation or production server changes.