Add strategy evolution and local loop runner

[sheepxux]

What changed

• add benchmark-gated task strategy evolution with versioned strategy and experiment artifacts
• add a local loopd runner with scheduling, atomic leases, dry-run and opt-in command executors, timeouts, and run artifacts
• add loopctl status/runs/pause/resume controls and safety validation
• add a Chinese-default README plus a complete English README

Why

This turns Loop-Engineering from a protocol-only project into a measurable self-improving loop contract plus a durable local runtime, while keeping the safety contract immutable.

Validation

• npm run smoke
• npm pack --dry-run
• Codex and Claude adapter skill validation
• manual loopd end-to-end dry-run

Summary by cubic

Adds strategy evolution and a local durable runner. v0.3.0 introduces loopd for scheduling and running loops, plus schemas and commands to manage strategies, experiments, and status.

• New Features

  • Local loopd start with --once or continuous polling; dry-run and opt‑in command executors.
  • Atomic file leases, command timeouts, durable run dirs, event logs, and expired-lease recovery.
  • New CLI: loopctl status, runs, pause, resume, and evolve.
  • Protocol updates: runner and evolution sections; state.paused; run-log.metrics; new strategy.schema.json and experiment.schema.json.
  • Validator checks for runner config and safe evolution (independent evaluator, low-risk for auto‑promotion).
  • Adapters honor LOOP_RUN_LOG (runner records the run) and read the active strategy.json when evolution is enabled.
  • New example: examples/self-improving-development/loop.yaml. Docs and templates updated. Package exposes loopd binary; version bumped to 0.3.0.

• Migration

  • Upgrade to v0.3.0. Run loopctl doctor.
  • To use the local runner, add runner to loop.yaml (e.g., executor: dry-run or command, workingDirectory, pollSeconds) and ensure schedule cadence is set. Validate with loopctl validate.
  • To enable strategy evolution, add evolution and set persistence.strategyPath and persistence.experimentDir; initialize the baseline strategy (re-init from templates or add strategy.json).
  • If you have a custom harness, write the run log to the LOOP_RUN_LOG path when present and do not call loopctl record from runner-managed commands.
  • Install/use the new loopd binary for local scheduling.

^{Written for commit 6bacd33. Summary will update on new commits.}