feat(upstream): custom 模型按媒体分类配置参考文件上限

- ModelConfig 新增 max_reference_videos / max_reference_audios
  (配合既有 max_reference_images);admin 读写接口透传
- custom 模型表单去掉「参考模式」下拉,改为参考图/参考视频/
  参考音频三个数量输入;合计张数走 totalRefLimit 校验
- vivid 上游 /v1/uploads 按文件魔数自动选 kind(image/video/
  audio)与 mime;openai 协议上游遇视频/音频参考自动跳过换号
- videoCaps 让 custom 模型的分类上限在画图台/文档徽标生效
- reference_mode 请求级 override 放行 "media"

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

backend/internal/service/admin_write.go

@@ -401,6 +401,8 @@ func (s *AdminWriteService) CreateModel(ctx context.Context, body map[string]any
 		DurationPricesAgent: jsonMap(body["duration_prices_agent"]),
 		Durations:           jsonArray(body["durations"]),
 		MaxReferenceImages:  intValue(body["max_reference_images"]),
+		MaxReferenceVideos:  intValue(body["max_reference_videos"]),
+		MaxReferenceAudios:  intValue(body["max_reference_audios"]),
 		ReferenceMode:       defaultString(strings.TrimSpace(stringValue(body["reference_mode"])), "none"),
 		Weight:              intValue(body["weight"]),
 		CreatedAt:           time.Now(),
@@ -477,6 +479,12 @@ func (s *AdminWriteService) UpdateModel(ctx context.Context, modelID string, bod
 	if _, ok := body["max_reference_images"]; ok {
 		patch["max_reference_images"] = intValue(body["max_reference_images"])
 	}
+	if _, ok := body["max_reference_videos"]; ok {
+		patch["max_reference_videos"] = intValue(body["max_reference_videos"])
+	}
+	if _, ok := body["max_reference_audios"]; ok {
+		patch["max_reference_audios"] = intValue(body["max_reference_audios"])
+	}
 	if _, ok := body["reference_mode"]; ok {
 		patch["reference_mode"] = defaultString(strings.TrimSpace(stringValue(body["reference_mode"])), "none")
 	}