fix(security): P0 安全修复 — 审计日志 PII 脱敏 + AI Token 计量 + backup.sh 拼写 + CI audit

1. 审计日志 PII 脱敏: audit_service.rs 中 old_value/new_value 自动 mask
   patient/consultation/follow_up 等资源类型的 PII 字段(id_number/phone/name 等)
2. AI Token 计量: chat_handler.rs 从 Provider response 和 AgentOrchestrator 提取
   实际 input_tokens/output_tokens，替代硬编码 0
3. AI display_hints: 从 AgentOrchestrator 传递 display_hints 给前端 ChatResponse
4. backup.sh: PGDATABSE 拼写错误修复为 PGDATABASE
5. CI: npm audit 移除 || true，高危漏洞阻止合并
6. 新增六维度深度分析报告 docs/discussions/2026-05-28

crates/erp-ai/src/handler/chat_handler.rs

@@ -240,6 +240,14 @@ where
     let provider_name = provider_arc.name().to_string();
     let supports_fc = provider_name != "ollama"; // Ollama generate_with_tools 未实现
 
+    // 收集 token 和 display_hints
+    #[allow(unused_assignments)]
+    let mut input_tokens: u32 = 0;
+    #[allow(unused_assignments)]
+    let mut output_tokens: u32 = 0;
+    let mut duration_ms: u64 = 0;
+    let mut collected_hints: Option<Vec<crate::agent::tool::DisplayHint>> = None;
+
     let result = if supports_fc {
         // FC provider：执行完整 Agent ReAct 循环
         let orchestrator = AgentOrchestrator::new(provider_arc, std::sync::Arc::new(registry));
@@ -256,6 +264,11 @@ where
                 tracing::error!(error = %e, "AI Agent run failed");
                 erp_core::error::AppError::Internal("AI 服务暂时不可用，请稍后再试".into())
             })?;
+        input_tokens = agent_result.total_input_tokens;
+        output_tokens = agent_result.total_output_tokens;
+        if !agent_result.display_hints.is_empty() {
+            collected_hints = Some(agent_result.display_hints);
+        }
         agent_result.reply
     } else {
         // 非 FC provider：降级为普通对话
@@ -279,6 +292,9 @@ where
                 tracing::error!(error = %e, "AI generate failed");
                 erp_core::error::AppError::Internal("AI 服务暂时不可用，请稍后再试".into())
             })?;
+        input_tokens = resp.input_tokens;
+        output_tokens = resp.output_tokens;
+        duration_ms = resp.duration_ms;
         resp.content
     };
 
@@ -297,7 +313,7 @@ where
         "AI chat response sent"
     );
 
-    // 记录用量的 token 消耗（简化模式下无法精确计量，记 0）
+    // 记录用量的 token 消耗
     if let Err(e) = ai_state
         .usage
         .log_usage(
@@ -305,9 +321,9 @@ where
             &provider_name,
             &run_params.model,
             "chat",
-            0,
-            0,
-            0,
+            input_tokens,
+            output_tokens,
+            duration_ms,
             0,
             false,
         )
@@ -362,7 +378,7 @@ where
         reply,
         message_id,
         iterations: if supports_fc { 1 } else { 0 },
-        display_hints: None,
+        display_hints: collected_hints,
     })))
 }