[Feature] Add ControlLoop primitive for convergence-style agent orchestration

[RagavRida]

Summary

I've implemented a ControlLoop primitive that ports the control-loop paradigm (thermostats, Kubernetes controllers, HumanLayer's agentic control-loop methodology) into HyperAgent as an opt-in, purely additive primitive.

Branch

👉 Ready for review: RagavRida:feat/control-loop (commit 0c0fb0f)
📊 Diff: Compare view — 1,542 additions, 3 deletions across 12 files

⚠️ I wasn't able to open a PR directly because this repo restricts PR creation to collaborators. Could a maintainer either add me as a collaborator or open the PR from my branch?

What's Included

• src/control-loop: ControlLoop class — a sensor measures the gap to a declared set point, a controller picks ONE item per iteration (FIFO default, customizable), an actuator applies the change via agent tasks
• Golden patterns + reviewer feedback (file re-read every iteration) rendered into actuator prompts via buildLoopGuidance
• Flow control: gates with bounded hold-polling (refuse new work while output awaits review), maxIterations, consecutive-failure caps
• Structural ControlLoopAgent contract keeps loops testable without a browser or LLM: scripts/test-control-loop.ts covers 13 deterministic scenarios
• Worked example: examples/control-loop/hn-worklist (HN digest worklist with pending-review gate and live-editable feedback file)
• Docs: README section, currentState.md, AGENTS.md updates

Merge-readiness

• ✅ Branches can be automatically merged (no conflicts)
• ✅ Purely additive — no breaking changes to existing code
• ✅ 13 deterministic test scenarios passing