fix(sync): prioritize bounded public coverage

[branarakic]

User impact

When a Core node has admitted a bounded batch of public Context Graphs, it now synchronizes both their finalized VM data and live SWM before spending the rest of the peer round on the potentially large agents/ontology bootstrap backlog.

Only scheduler-vetted automatic public CGs receive this priority. Configured or private CGs remain after the Agents/Ontology bootstrap, so private SWM authorization and key recovery always use a refreshed Agents registry. Edge opt-in behavior is unchanged.

Problem

The previous peer round combined system graphs and the bounded public batch into one durable request, then deferred all SWM until that combined request completed. A large or pressured agents graph could therefore time out first and terminate the round even though the user-facing public CGs were ready to converge.

The formal M1 gate reproduced this: the Core spent its round transferring tens of thousands of system triples and hit store/RPC pressure before emitting an automatic coverage entry for the three fresh public graphs.

Before

sequenceDiagram
    participant C as "Core"
    participant P as "Peer"

    C->>P: Durable sync agents + ontology + public/configured CGs
    P-->>C: Large agents backlog
    P--xC: Timeout or backpressure
    Note over C: Round stops
    Note over C: Public VM incomplete and public SWM never attempted

Loading

After

sequenceDiagram
    participant C as "Core"
    participant P as "Peer"

    C->>P: Durable sync scheduler-vetted public batch
    P-->>C: Public VM data
    C->>P: Sync public SWM for the same batch
    P-->>C: Public SWM data
    C->>P: Bootstrap agents + ontology + configured scope
    P-->>C: Refreshed registry and configured VM data
    C->>P: Sync configured/private SWM
    Note over C: Public VM and SWM are already evidenced
    Note over C: Private SWM runs only after registry refresh

Loading

Implementation

• Define one canonical SyncOnConnectScopePlan; the base and lifecycle plans extend it instead of duplicating the contract.
• Pass only scheduler-vetted automatic public CGs through preBootstrapPublicContextGraphIds.
• Execute durable and shared phases through one ordered helper with common accounting, metadata refresh, and stop-on-pressure behavior.
• Preserve the legacy empty initial probe for Edge accounting compatibility.
• Avoid repeating SWM work for graphs completed by the early public phase.
• Keep configured/private graphs in the post-bootstrap path.

Validation

• 139 focused tests passed across:
  • sync-on-connect-retry.test.ts
  • sync-on-connect-churn.test.ts
  • core-public-coverage-scheduler.test.ts
  • discovery-subscription-boundary.test.ts
  • sync-coverage-evidence-runtime.test.ts
  • sync-coverage-evidence-edge-periodic.test.ts
• Includes an exact phase-order regression proving private SWM is attempted only after Agents/Ontology and metadata refresh.
• pnpm --filter @origintrail-official/dkg-agent run build
  • TypeScript, type tests, and package-root test passed.
• git diff --check
  • passed.

Stack

• Base: fix(sync): wake Core coverage after scope expansion #2046 (codex/rfc64-m1-core-discovery-wakeup)
• This PR is the next independently reviewable M1 layer.
• It addresses the scheduling blocker exposed by the formal three-node gate; the gate must still be rerun on the exact reviewed stack before M1 can pass.