feat: OpenClaw integration runner (structured callback + state + docs)

[sqrtqiezi]

Summary

This PR implements an MVP for issue #909 (OpenClaw integration):

• Add tools/openclaw_runner.py as a stable OpenClaw execution entrypoint
• Add structured JSON callback result for each run (status/duration/error_type/summary/latest_db/checksum/state_file/log_file)
• Add state persistence (output/openclaw/openclaw_state.json) with idempotency hints:
  • last_sent_at
  • latest_db
  • latest_db_mtime
  • latest_db_checksum
  • changed_since_last
• Add failure classification (fetch_failed, db_missing, notify_failed, etc.) and fallback hint support (--fallback-last-on-failure) to avoid silent failure
• Add integration doc: docs/openclaw-integration.md

Why

OpenClaw-side automation needs a stable entrypoint and machine-readable run callback for scheduling, observability and degraded alerting.

Usage

python tools/openclaw_runner.py --run-mode scheduled
python tools/openclaw_runner.py --run-mode manual
python tools/openclaw_runner.py --run-mode scheduled --fallback-last-on-failure

Verification

• Dry-run output validates structured callback and state transitions:
  • first run: changed_since_last=true
  • subsequent run with same DB: changed_since_last=false

Related: #909