[Public Agent Network]: Keeper runtime automation and replay-safe lifecycle jobs

[FlowmemoryAI]

Goal

Add the keeper/runtime automation layer required to operate public agents beyond static launch projections.

Scope

• Fuel metering loop
• Memory receipt and correction projection
• Challenge/correction timelines
• Replay-safe job orchestration
• Operator runbook and failure drill

Acceptance Criteria

• Keeper loop consumes explicit local/test inputs and emits deterministic public-safe state.
• Fuel debit and receipt anchoring paths are covered by tests.
• Correction and challenge status project through control-plane, SDK, and dashboard surfaces.
• Failure/replay behavior is documented with an operator drill.

Verification

• Unit/integration tests cover stale sequence, replay, correction, and failed job branches.
• Local e2e produces deterministic outputs without secrets.