feat(settings): M3 全局设置 + voice 动态路由 + env 收敛

M3 完成:env 业务配置全进 DB,WebUI 可改,改完立即生效(无需重启)。

新增:
- controller/settings.go: 3 个 handler
  - GET /api/settings: 读所有设置,api_key 永远打码
  - PUT /api/settings: 部分更新(白名单字段);default_speaker 改时校验
    音色必须在 voices 表里(避免引用不存在);改完调 LoadRuntimeConfig 即时生效
  - PUT /api/settings/api-key: 独立端点,只改 api_key

改造:
- setting/config.go: 用 LoadRuntimeConfig(store) 替代 InitTTSConfig
  - 不再读 BYTEDANCE_TTS_* env,完全从 store.Settings 拿
  - 必填项缺失返 error,设 TTSConfigErr(/v1/audio/speech 返 503)
  - 定义 Store 接口(避免 setting → store 循环 import)
  - 字段映射: BYTEDANCE_TTS_API_KEY → api_key, BYTEDANCE_TTS_RESOURCE_ID
    → default_resource_id, ... 全部一一对应
- controller/tts.go: voice 路由
  - voice 为空 → 用 setting.TTSOptions.Speaker (即 store.default_speaker)
  - voice 非空 → 查 voices 表,替换 opts.Speaker / ResourceID / Model
  - 命中但 enabled=0 → 仍可用(显式 voice 覆盖)
  - 未命中 → 400 unknown_voice
  - voice 路由日志: "[tts] voice=xxx 命中 (speaker=... resource=... model=...)"
- router/admin.html: 加 #settings tab (24.78KB, 仍 <30KB 预算)
  - API key 区:打码显示 + 独立更新按钮
  - 路由/格式区:default_resource_id / default_speaker (下拉源 /api/voices) /
    default_format / sample_rate / model
  - "保存" 按钮 PUT /api/settings + 立即 reload (settingsOk 提示)
- router/router.go: 挂 /api/settings[/{api-key}] 三路由
- main.go: Detect 之后调 LoadRuntimeConfig,失败也允许进入
  (env 阶段无 db,LogStartupSummary 仍显示 BYTEDANCE_TTS_* 必填项用于诊断)

测试(都被 .gitignore 排除,本地保留):
- 已有 store/installer/middleware/router/telemetry 测试保持全绿
- e2e 验证(本机跑通 9 场景):
  - install 写入 3 voices + 7 settings
  - 重启后 main.go log: "TTS 运行时配置已加载(api_key=***, speaker=default-voice, resource=volc.megatts.icl, format=wav)" — 确认走 DB 不读 env
  - GET /api/settings 返所有字段
  - voice=clone-voice-1 命中 S_clone_1 + seed-tts-2.0-expressive
  - voice=clone-voice-2 命中 S_clone_2 + volc.megatts.icl-2
  - voice=unknown 返 400 unknown_voice
  - PUT /api/settings 改 default_speaker 即时生效
  - PUT /api/settings/api-key 改 key 即时生效
  - PUT /api/settings default_speaker=ghost 返 400 default_speaker_missing

注意: e2e 里的 TTS 500 是因为我用假 key 调上游,路由逻辑本身正确。
换真火山 API key 即可 work。

二进制大小: ~17MB → ~18MB (+1MB)

未 push(等用户测试 M1+M2+speaker fix 全部稳定后,跟 51bd7ae+70865a3 一起或单独 push)

controller/tts.go

@@ -18,6 +18,7 @@ import (
 	"github.com/volcano-tts/tts-api/metrics"
 	"github.com/volcano-tts/tts-api/middleware"
 	"github.com/volcano-tts/tts-api/setting"
+	"github.com/volcano-tts/tts-api/store"
 	"github.com/volcano-tts/tts-api/telemetry"
 	"github.com/volcano-tts/tts-api/version"
 )
@@ -153,6 +154,44 @@ func OpenaiTTSHandler(w http.ResponseWriter, r *http.Request) {
 	opts := setting.TTSOptions
 	opts.Text = req.Input
 
+	// M3: voice 路由
+	//   - voice 为空 → 用 setting.TTSOptions.Speaker (即 store 里的 default_speaker)
+	//   - voice 非空 → 查 voices 表,替换 opts.Speaker / ResourceID / Model
+	//   - 命中但 enabled=0 → 仍可用(用户显式传 voice 即覆盖 enabled 状态;若想禁用在 admin UI 关掉就行)
+	//   - 未命中 → 400 "unknown voice: <name>"
+	if req.Voice != "" {
+		s := GetAdminStore()
+		if s == nil {
+			log.Printf("警告: voice=%s 路由但 store 未初始化 - 路径=%s", req.Voice, r.URL.Path)
+			middleware.SendJSONError(w, http.StatusServiceUnavailable,
+				"voice routing requires database; not initialized",
+				"configuration_error", "db_not_ready")
+			return
+		}
+		v, err := s.VoiceGetByName(req.Voice)
+		if err != nil {
+			if err == store.ErrNotFound {
+				log.Printf("警告: 未知 voice=%s - 路径=%s 客户端=%s", req.Voice, r.URL.Path, middleware.GetClientIP(r))
+				middleware.SendJSONError(w, http.StatusBadRequest,
+					fmt.Sprintf("unknown voice: %s", req.Voice),
+					"invalid_request_error", "unknown_voice")
+				return
+			}
+			log.Printf("警告: voice 查库失败 - 错误=%v voice=%s", err, req.Voice)
+			middleware.SendJSONError(w, http.StatusInternalServerError,
+				"voice lookup failed", "server_error", "db_read_failed")
+			return
+		}
+		// 覆盖 opts(API key / UID 保留自 setting.TTSOptions)
+		opts.Speaker = v.Speaker
+		opts.ResourceID = v.ResourceID
+		if v.Model != "" {
+			opts.Model = v.Model
+		}
+		log.Printf("[tts] voice=%s 命中 (speaker=%s resource=%s model=%s) - 客户端=%s",
+			req.Voice, v.Speaker, v.ResourceID, v.Model, middleware.GetClientIP(r))
+	}
+
 	ctx, cancel := context.WithTimeout(r.Context(), setting.TTSTimeout)
 	defer cancel()