refactor(monitoring): N2 - tach monitoring_tab.py, CanonicalAuditLogger, MonitoringQueryService, go circular import, sandbox matrix

- ui/monitoring_tab.py (1546 dong) tach thanh presentation/monitoring/**
  (container + 7 tab/card + shared helper), ui/monitoring_tab.py con lai
  re-export shim de app.py khong doi.
- infrastructure/telemetry/audit_logger.py: CanonicalAuditLogger, core/audit_log.py
  thanh wrapper mong, tuong thich nguoc 100% voi schema .jsonl cu.
- application/monitoring/monitoring_query_service.py: MonitoringQueryService
  read-only, filter/sort/pagination, khong import PySide6.
- Go circular import model_pricing<->usage_tracker va agent_security<->
  agent_security_alert (core/agent_security_types.py moi).
- infrastructure/sandbox/sandbox_capabilities.py: SandboxCapabilityMatrix
  theo OS (Windows/Linux/macOS), chua dau noi vao core/sandbox_manager.py.
- conftest.py: sua loi checkout khong ten cowork_local khien pytest import
  nham thu muc khac.
- 77 test moi, 167/167 pass. QA da xac nhan UI/business logic khong doi
  (xem evidence/report/unified_report.html).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

core/agent_security.py

@@ -27,27 +27,12 @@ from __future__ import annotations
 
 import json
 import re
-from dataclasses import dataclass
 from typing import List, Optional
 
 from ..providers.base import Provider
 from . import security_rules
-
-
-class SecurityBlocked(RuntimeError):
-    """A guardrail refused an action. ``verdict`` carries the full detail for
-    the admin alert; ``str(exc)`` is the short, user-facing reason."""
-
-    def __init__(self, verdict: "SecurityVerdict"):
-        super().__init__(verdict.reason or f"Blocked by agent security ({verdict.layer}).")
-        self.verdict = verdict
-
-
-@dataclass
-class SecurityVerdict:
-    allowed: bool
-    reason: str = ""
-    layer: str = ""   # "prompt" | "attachment" | "command"
+from .agent_security_alert import notify_admin
+from .agent_security_types import SecurityBlocked, SecurityVerdict
 
 
 def combined_rules_text(config, max_chars: int = 8000, agent_kind: str = "cowork") -> str:
@@ -236,7 +221,6 @@ def enforce_prompt(provider: Provider, messages: List[dict], config, emit,
     emit({"type": "notice", "level": "warning",
           "text": f"🛡 Yêu cầu bị chặn bởi Agent Security: {verdict.reason}"})
     from . import audit_log
-    from .agent_security_alert import notify_admin
 
     audit_log.record("security_block", "prompt", False, verdict.reason)
     notify_admin(config, verdict, detail=user_text[:1000])
@@ -266,7 +250,6 @@ def enforce_command(provider: Provider, name: str, args: dict, config, emit,
     emit({"type": "notice", "level": "warning",
           "text": f"🛡 Lệnh bị chặn bởi Agent Security ({verdict.layer}): {verdict.reason}"})
     from . import audit_log
-    from .agent_security_alert import notify_admin
 
     audit_log.record("security_block", name, False, f"{verdict.layer}: {verdict.reason}")
     notify_admin(config, verdict, detail=command)