feat: VLM 模型切换 qwen3-vl-plus + Diet Agent 患者档案联动

- VLM 模型切换为 qwen3-vl-plus（中餐识别准确率大幅提升）
- VLM Prompt 简化为仅识别食物名+份量+热量
- 营养分析/禁忌提醒移至 Diet Agent（可查患者档案）
- Diet Agent Prompt 强化：过敏→红色警告，低盐低脂→黄色提醒
- 上传限制调整至 20MB
- 服务端图片压缩参数优化（960px/Q72）

backend/src/Health.Infrastructure/AI/ai_clients.cs

@@ -85,12 +85,12 @@ public sealed class DeepSeekClient(HttpClient http, IConfiguration config)
 }
 
 /// <summary>
-/// 千问 VL 视觉客户端（食物识别 + 报告解读）
+/// VLM 视觉客户端——支持千问/豆包，通过 .env 切换
 /// </summary>
-public sealed class QwenVisionClient(HttpClient http, IConfiguration config)
+public sealed class VisionClient(HttpClient http, IConfiguration config)
 {
     private readonly HttpClient _http = http;
-    private readonly string _model = config["QWEN_VISION_MODEL"] ?? "qwen-vl-max";
+    private readonly string _model = config["VLM_MODEL"] ?? "doubao-vision-pro";
     private readonly JsonSerializerOptions _jsonOptions = new()
     {
         PropertyNamingPolicy = JsonNamingPolicy.SnakeCaseLower,