feat: Security hardening, CLI refactoring, and agent metrics

[foozio]

This PR introduces several requested improvements:

• P0: Hardened rate limiting and added execution guardrails for process/shell tools. Addressed panic-prone unwrap cases and redacted sensitive logs.
• P1: Refactored CLI into command modules for better maintainability.
• P2: Enhanced RAG with BM25 ranking, added history pruning for long-running deployments, and execution metrics. Added documentation for production security baselines.

[volumeee]

🔍 Review untuk PR #1 - Security & Architecture Check

Terima kasih atas PR ini, @foozio! Secara keseluruhan ini adalah improvement yang solid.
Saya punya beberapa pertanyaan untuk memastikan security hardening-nya solid:

1️⃣ ZENCLAW_ALLOW_COMMAND_EXEC Enforcement

Status: ✅ Good - tapi need clarification

Saya lihat policy ini ditambahkan di Commit b56b5bf. Pertanyaan:

• Apakah enforcement-nya default DENY (safe by default) atau default ALLOW?
• Jika env var ZENCLAW_ALLOW_COMMAND_EXEC tidak diset, sistem akan:
  • Reject semua command execution (preferred)
  • Allow execution dengan warning
  • Something else?

Bisa tunjukkan code snippet enforcement-nya?

2️⃣ Rate Limiting & IP Spoofing Risk

Status: ⚠️ Needs Clarification

Di crates/zenclaw-hub/src/middleware.rs, rate limiting key pakai:

let key = headers
    .get("x-forwarded-for")
    .and_then(|v| v.to_str().ok())
    .unwrap_or("unknown")
    .to_string();