fix: idempotent register guard + governance detail logging

[jlin53882]

Summary

Two fixes for memory-lancedb-pro plugin:

1. Idempotent register() guard

Problem: During gateway boot, OpenClaw calls register() multiple times (once per agent session). Without a guard, the expensive initialization (MemoryStore, embedder, hooks) runs every time, causing redundant initialization blocks in logs.

Fix: Add let _initialized = false module-level flag. On repeated register() calls, log a debug message and return early.

let _initialized = false;

register(api: OpenClawPluginApi) {
  if (_initialized) {
    api.logger.debug("memory-lancedb-pro: register() called again — skipping re-init (idempotent)");
    return;
  }
  _initialized = true;
  // ...
}

Export _resetInitialized() for test harness reset.

2. Governance filter observability

Problem: Governance filter decisions were only logged as aggregate counts (stateFiltered=N), making it impossible to trace which entries were filtered and why without replaying.

Fix: Add per-entry debug log with id, reason, score, and text snippet:

api.logger.debug(`memory-lancedb-pro: governance: filtered id=${r.entry.id} reason=state(${meta.state}) score=${r.score?.toFixed(3)} text=${r.entry.text.slice(0, 50)}`);

Testing

• Regression tests: ✅ All pass (All regression tests passed!)
• Local extensions: ✅ Gateway starts with initialized successfully only once

Files changed

• index.ts: +14 lines (guard + logging + export)