refactor: 将客服后端重构为宿主可嵌入模块

- 注入数据库、运行时配置、统一响应、文件存储和平台 AI 能力，补充业务读写工具与客户快捷操作契约。

- 移除模块内重复的组织、客户、工单、标签、技能、旧工作流、MCP 和迁移实现，将身份权限与业务主体交由宿主管理。

- 使用 libSQL 重构向量存储，并完善图片消息、访客身份、排队调度、企业微信和支持聊天页面。

- 统一 HTTP、DTO 与 WebSocket 的 snake_case 协议，补齐模块初始化、业务动作和公共载荷等回归测试。

internal/ai/tooling/registry.go

@@ -32,14 +32,13 @@ type Definition struct {
 // Policy is supplied by the caller's agent/runtime context for one invocation.
 // An empty AllowedToolCodes means the caller did not impose an allow-list.
 type Policy struct {
-	AllowedToolCodes      []string
-	SkillAllowedToolCodes []string
-	AllowedRiskLevels     []string
-	CallCount             int
-	TotalCallCount        int
-	MaxTotalCalls         int
-	MaxArgumentBytes      int
-	Confirmed             bool
+	AllowedToolCodes  []string
+	AllowedRiskLevels []string
+	CallCount         int
+	TotalCallCount    int
+	MaxTotalCalls     int
+	MaxArgumentBytes  int
+	Confirmed         bool
 }
 
 type Invocation struct {
@@ -69,24 +68,7 @@ func (r *Registry) Resolve(toolCode string) (Definition, error) {
 	if spec, ok := toolx.GetRegisteredToolSpec(toolCode); ok {
 		return definitionFromSpec(spec), nil
 	}
-	serverCode, toolName := toolx.SplitMCPToolCode(toolCode)
-	if serverCode == "" || toolName == "" {
-		return Definition{}, fmt.Errorf("unsupported tool code: %s", toolCode)
-	}
-	// MCP tools are explicitly selected by an administrator before an Agent can
-	// call them. Treat that persisted allow-list as the authorization boundary;
-	// only tools with an explicit built-in policy require extra confirmation.
-	return Definition{
-		Code:                toolCode,
-		Name:                toolName,
-		InputSchema:         map[string]any{"type": "object", "additionalProperties": true},
-		SourceType:          enums.ToolSourceTypeMCP,
-		RiskLevel:           RiskLevelRead,
-		RequireConfirmation: false,
-		MaxCallsPerRun:      3,
-		TimeoutMS:           30000,
-		IdempotencyMode:     "caller",
-	}, nil
+	return Definition{}, fmt.Errorf("unsupported tool code: %s", toolCode)
 }
 
 func (r *Registry) Authorize(definition Definition, policy Policy) error {
@@ -102,9 +84,6 @@ func (g *PolicyGuard) Authorize(invocation Invocation) error {
 	if len(policy.AllowedToolCodes) > 0 && !containsCanonicalToolCode(policy.AllowedToolCodes, definition.Code) {
 		return fmt.Errorf("tool is not allowed: %s", definition.Code)
 	}
-	if len(policy.SkillAllowedToolCodes) > 0 && !containsCanonicalToolCode(policy.SkillAllowedToolCodes, definition.Code) {
-		return fmt.Errorf("tool is not allowed by the selected skill: %s", definition.Code)
-	}
 	if len(policy.AllowedRiskLevels) > 0 && !containsString(policy.AllowedRiskLevels, definition.RiskLevel) {
 		return fmt.Errorf("tool risk level is not allowed: %s", definition.RiskLevel)
 	}
@@ -147,37 +126,18 @@ func definitionFromSpec(spec toolx.ToolSpec) Definition {
 		definition.InputSchema = requiredObjectSchema([]string{"query"}, map[string]any{"query": map[string]any{"type": "string"}})
 	case toolx.GraphTriageServiceRequest.Code:
 		definition.InputSchema = objectSchema(map[string]any{
-			"goal":              map[string]any{"type": "string"},
-			"observedIssue":     map[string]any{"type": "string"},
-			"needTicket":        map[string]any{"type": "boolean"},
-			"needHumanHandoff":  map[string]any{"type": "boolean"},
-			"additionalContext": map[string]any{"type": "string"},
+			"goal":               map[string]any{"type": "string"},
+			"observed_issue":     map[string]any{"type": "string"},
+			"need_human_handoff": map[string]any{"type": "boolean"},
+			"additional_context": map[string]any{"type": "string"},
 		})
 	case toolx.GraphAnalyzeConversation.Code:
 		definition.InputSchema = objectSchema(map[string]any{
-			"goal":              map[string]any{"type": "string"},
-			"observedIssue":     map[string]any{"type": "string"},
-			"needTicket":        map[string]any{"type": "boolean"},
-			"needHumanHandoff":  map[string]any{"type": "boolean"},
-			"needQualityCheck":  map[string]any{"type": "boolean"},
-			"additionalContext": map[string]any{"type": "string"},
-		})
-	case toolx.GraphPrepareTicketDraft.Code:
-		definition.InputSchema = objectSchema(map[string]any{
-			"title":           map[string]any{"type": "string"},
-			"description":     map[string]any{"type": "string"},
-			"issue":           map[string]any{"type": "string"},
-			"impact":          map[string]any{"type": "string"},
-			"expectedOutcome": map[string]any{"type": "string"},
-			"currentAttempt":  map[string]any{"type": "string"},
-		})
-	case toolx.GraphCreateTicketConfirm.Code:
-		definition.RiskLevel = RiskLevelWrite
-		definition.RequireConfirmation = true
-		definition.MaxCallsPerRun = 1
-		definition.IdempotencyMode = "business"
-		definition.InputSchema = requiredObjectSchema([]string{"title", "description"}, map[string]any{
-			"title": map[string]any{"type": "string"}, "description": map[string]any{"type": "string"},
+			"goal":               map[string]any{"type": "string"},
+			"observed_issue":     map[string]any{"type": "string"},
+			"need_human_handoff": map[string]any{"type": "boolean"},
+			"need_quality_check": map[string]any{"type": "boolean"},
+			"additional_context": map[string]any{"type": "string"},
 		})
 	case toolx.GraphHandoffConversation.Code:
 		definition.RiskLevel = RiskLevelWrite