architecture: phased migration plan to libevent orchestration

[somethingwithproof]

Summary

Outline and execute a safe, phased migration to a libevent-based orchestration layer in Spine. The near-term goal is improved timer/signal/process supervision reliability without rewriting all polling logic at once.

Work

• Define target architecture: single libevent loop for orchestration + existing worker threads for blocking SNMP/DB operations.
• Add configure/build detection for libevent and keep non-libevent build path available.
• Implement migration in phases:
  • Phase 1: bootstrap event loop lifecycle (startup/shutdown/error paths).
  • Phase 2: migrate timers and signal handling to libevent primitives.
  • Phase 3: move child-process supervision (timeouts/output caps/cleanup) under event-loop control.
  • Phase 4: integrate queue wakeups/notifications between worker threads and event loop.
  • Phase 5: evaluate optional async conversions where dependencies support it; keep blocking paths isolated until proven safe.
• Add feature gating so rollout can be enabled incrementally.
• Add regression/perf validation for poll cycle correctness, timeout behavior, and shutdown determinism.
• Update operator/developer docs with architecture and rollout guidance.

Acceptance Criteria

• Spine builds and runs both with and without libevent support.
• Event-loop path preserves existing poll results and failure semantics.
• Timer/signal/child-process handling is centralized in the event loop and verified by tests.
• Rollout is gated and reversible without code rollback.
• Documentation describes architecture, flags, and migration/rollback steps.