feat: 监控告警体系统一重构 + 自动化任务调度引擎 + 监控大屏拆分

告警中心
- 新增统一告警(alert_unified)、监控告警规则(monitor_alert_rule)、指标目录(alert_metric_catalog)
- 通知策略与消息分发(alert_notify_policy/dispatch/message)、静默规则
- 告警引擎按阈值/SNMP 拆分(alert_engine_monitor/snmp)

监控采集
- SNMP 常量/模板与历史指标(snmp templates/constants、snmp_metrics_history、snmp_collect)
- 中间件监控 Redis 采集与种子(middleware_monitor_redis/seed)
- 站点探测 pkg(siteprobe) + 站点视图(site_view)

自动化任务
- cron 调度器(cronscheduler)、脚本运行器(scriptrunner)、系统任务(systask)及运行记录(cron_job_run、system_task)

审计与凭证
- 审计设置(audit_setting) + 历史清理(auditcleaner)

前端
- 监控页面(host/middleware/site/snmp)拆分为子模块并大屏化
- 告警中心多 Tab(规则/指标目录/通知策略/静默)、任务管理组件重构、Agent CLI Markdown 渲染
- 移除旧脚本卡片/详情/日志组件，统一为 ScriptTable/ScriptCodeEditor/TerminalOutput

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

backend/controller/agent_cli_history.go

@@ -0,0 +1,89 @@
+// agent_cli_history.go - Agent CLI 历史会话恢复为 LLM 对话上下文 // Logan.Li
+package controller
+
+import (
+	"encoding/json"
+	"fmt"
+
+	"ezops/model"
+	"ezops/pkg/agentcli/llm"
+)
+
+// buildConvHistoryFromMessages 将 DB 消息重建为追问用的 prior 历史（不含 system） // Logan.Li
+func buildConvHistoryFromMessages(msgs []model.AgentCLIMessage) []llm.Message {
+	var out []llm.Message
+	var pendingThought string
+	var pendingToolCalls []llm.ToolCall
+	var pendingToolCallIDs []string
+
+	flushAssistant := func() {
+		if pendingThought == "" && len(pendingToolCalls) == 0 {
+			return
+		}
+		out = append(out, llm.Message{
+			Role:      "assistant",
+			Content:   pendingThought,
+			ToolCalls: append([]llm.ToolCall(nil), pendingToolCalls...),
+		})
+		for _, tc := range pendingToolCalls {
+			pendingToolCallIDs = append(pendingToolCallIDs, tc.ID)
+		}
+		pendingThought = ""
+		pendingToolCalls = nil
+	}
+
+	for _, m := range msgs {
+		switch m.Type {
+		case "user":
+			flushAssistant()
+			out = append(out, llm.Message{Role: "user", Content: m.Content})
+		case "thought":
+			pendingThought = m.Content
+		case "tool_call":
+			args := validateToolArgsJSON(m.ToolArgs)
+			tcID := fmt.Sprintf("hist-%d", m.ID)
+			pendingToolCalls = append(pendingToolCalls, llm.ToolCall{
+				ID: tcID, Type: "function",
+				Function: llm.FunctionCall{Name: m.ToolName, Arguments: args},
+			})
+		case "tool_result":
+			if pendingThought != "" || len(pendingToolCalls) > 0 {
+				flushAssistant()
+			}
+			toolID := ""
+			if len(pendingToolCallIDs) > 0 {
+				toolID = pendingToolCallIDs[0]
+				pendingToolCallIDs = pendingToolCallIDs[1:]
+			}
+			content := m.Content
+			if content == "" {
+				content = m.ToolOutput
+			}
+			out = append(out, llm.Message{
+				Role: "tool", Name: m.ToolName, Content: content, ToolCallID: toolID,
+			})
+		case "done":
+			flushAssistant()
+			out = append(out, llm.Message{Role: "assistant", Content: m.Content})
+		case "error":
+			flushAssistant()
+			out = append(out, llm.Message{Role: "assistant", Content: "错误: " + m.Content})
+		default:
+			// 忽略 progress/session 等 // Logan.Li
+		}
+	}
+	flushAssistant()
+	return out
+}
+
+// validateToolArgsJSON 校验工具参数 JSON，非法则回退空对象 // Logan.Li
+func validateToolArgsJSON(raw string) string {
+	if raw == "" {
+		return "{}"
+	}
+	var tmp map[string]interface{}
+	if json.Unmarshal([]byte(raw), &tmp) != nil {
+		return "{}"
+	}
+	return raw
+}