refactor: 后端代码质量优化 - 复用公共库、使用标准库、类型安全错误判断

## 高优先级修复
- stats_service_impl: 使用 strings.SplitN 替代错误的索引分割
- provider_handler: 使用 errors.Is(err, gorm.ErrDuplicatedKey) 替代字符串匹配
- client: 重写 isNetworkError 使用 errors.As/Is 类型安全判断
- proxy_handler: 使用 encoding/json 标准库解析 JSON（extractModelName、isStreamRequest）

## 中优先级修复
- stats_handler: 添加 parseDateParam 辅助函数消除重复日期解析
- pkg/errors: 新增 ErrRequestCreate/Send/ResponseRead 错误类型和 WithCause 方法
- client: 使用结构化错误替代 fmt.Errorf
- ConversionEngine: logger 依赖注入，替换所有 zap.L() 调用

## 低优先级修复
- encoder: 删除 joinStrings，使用 strings.Join
- adapter: 删除 modelInfoRegex 正则，使用 isModelInfoPath 字符串函数

## 文档更新
- README.md: 添加公共库使用指南和编码规范章节
- specs: 同步 delta specs 到 main specs（error-handling、structured-logging、request-validation）

## 归档
- openspec/changes/archive/2026-04-20-refactor-backend-code-quality/

backend/internal/conversion/openai/adapter.go

@@ -2,7 +2,7 @@ package openai
 
 import (
 	"encoding/json"
-	"regexp"
+	"strings"
 
 	"nex/backend/internal/conversion"
 	"nex/backend/internal/conversion/canonical"
@@ -16,8 +16,6 @@ func NewAdapter() *Adapter {
 	return &Adapter{}
 }
 
-var modelInfoRegex = regexp.MustCompile(`^/v1/models/[^/]+$`)
-
 // ProtocolName 返回协议名称
 func (a *Adapter) ProtocolName() string { return "openai" }
 
@@ -34,7 +32,7 @@ func (a *Adapter) DetectInterfaceType(nativePath string) conversion.InterfaceTyp
 		return conversion.InterfaceTypeChat
 	case nativePath == "/v1/models":
 		return conversion.InterfaceTypeModels
-	case modelInfoRegex.MatchString(nativePath):
+	case isModelInfoPath(nativePath):
 		return conversion.InterfaceTypeModelInfo
 	case nativePath == "/v1/embeddings":
 		return conversion.InterfaceTypeEmbeddings
@@ -45,6 +43,15 @@ func (a *Adapter) DetectInterfaceType(nativePath string) conversion.InterfaceTyp
 	}
 }
 
+// isModelInfoPath 判断是否为模型详情路径（/v1/models/{id}）
+func isModelInfoPath(path string) bool {
+	if !strings.HasPrefix(path, "/v1/models/") {
+		return false
+	}
+	suffix := path[len("/v1/models/"):]
+	return suffix != "" && !strings.Contains(suffix, "/")
+}
+
 // BuildUrl 根据接口类型构建 URL
 func (a *Adapter) BuildUrl(nativePath string, interfaceType conversion.InterfaceType) string {
 	switch interfaceType {

backend/internal/conversion/openai/adapter_test.go

@@ -112,6 +112,28 @@ func TestAdapter_SupportsInterface(t *testing.T) {
 	}
 }
 
+func TestIsModelInfoPath(t *testing.T) {
+	tests := []struct {
+		name     string
+		path     string
+		expected bool
+	}{
+		{"model_info", "/v1/models/gpt-4", true},
+		{"model_info_with_dots", "/v1/models/gpt-4.1-preview", true},
+		{"models_list", "/v1/models", false},
+		{"nested_path", "/v1/models/gpt-4/versions", false},
+		{"empty_suffix", "/v1/models/", false},
+		{"unrelated", "/v1/chat/completions", false},
+		{"partial_prefix", "/v1/model", false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.expected, isModelInfoPath(tt.path))
+		})
+	}
+}
+
 func TestAdapter_EncodeError_InvalidInput(t *testing.T) {
 	a := NewAdapter()
 	convErr := conversion.NewConversionError(conversion.ErrorCodeInvalidInput, "参数无效")

backend/internal/conversion/openai/encoder.go

@@ -2,6 +2,7 @@ package openai
 
 import (
 	"encoding/json"
+	"strings"
 	"time"
 
 	"nex/backend/internal/conversion"
@@ -89,7 +90,7 @@ func encodeSystemAndMessages(req *canonical.CanonicalRequest) []map[string]any {
 		for _, b := range v {
 			parts = append(parts, b.Text)
 		}
-		text := joinStrings(parts, "\n\n")
+		text := strings.Join(parts, "\n\n")
 		if text != "" {
 			messages = append(messages, map[string]any{
 				"role":    "system",
@@ -132,7 +133,7 @@ func encodeMessage(msg canonical.CanonicalMessage) []map[string]any {
 
 		if len(toolUses) > 0 {
 			if len(textParts) > 0 {
-				m["content"] = joinStrings(textParts, "")
+				m["content"] = strings.Join(textParts, "")
 			} else {
 				m["content"] = nil
 			}
@@ -149,7 +150,7 @@ func encodeMessage(msg canonical.CanonicalMessage) []map[string]any {
 			}
 			m["tool_calls"] = tcs
 		} else if len(textParts) > 0 {
-			m["content"] = joinStrings(textParts, "")
+			m["content"] = strings.Join(textParts, "")
 		} else {
 			m["content"] = ""
 		}
@@ -286,7 +287,7 @@ func encodeResponse(resp *canonical.CanonicalResponse) ([]byte, error) {
 	message := map[string]any{"role": "assistant"}
 	if len(toolUses) > 0 {
 		if len(textParts) > 0 {
-			message["content"] = joinStrings(textParts, "")
+			message["content"] = strings.Join(textParts, "")
 		} else {
 			message["content"] = nil
 		}
@@ -303,13 +304,13 @@ func encodeResponse(resp *canonical.CanonicalResponse) ([]byte, error) {
 		}
 		message["tool_calls"] = tcs
 	} else if len(textParts) > 0 {
-		message["content"] = joinStrings(textParts, "")
+		message["content"] = strings.Join(textParts, "")
 	} else {
 		message["content"] = ""
 	}
 
 	if len(thinkingParts) > 0 {
-		message["reasoning_content"] = joinStrings(thinkingParts, "")
+		message["reasoning_content"] = strings.Join(thinkingParts, "")
 	}
 
 	var finishReason *string
@@ -488,18 +489,6 @@ func encodeRerankResponse(resp *canonical.CanonicalRerankResponse) ([]byte, erro
 	})
 }
 
-// joinStrings 拼接字符串切片
-func joinStrings(parts []string, sep string) string {
-	result := ""
-	for i, p := range parts {
-		if i > 0 {
-			result += sep
-		}
-		result += p
-	}
-	return result
-}
-
 // mergeConsecutiveRoles 合并连续同角色消息（拼接内容）
 func mergeConsecutiveRoles(messages []map[string]any) []map[string]any {
 	if len(messages) <= 1 {