fix(llm): 分场景使用不同模型

- nvidia 激活、模型改为 step-3.7-flash → 用于日常生成/采集/优化等
- compliance_optimizer.py 显式指定 provider=sensenova → 合规审查走 deepseek-v4-flash
- _FALLBACK 同步更新为 step-3.7-flash
- 管理后台标签同步更新

platform/backend/app/core/nvidia_client.py

@@ -30,7 +30,7 @@ _API_KEYS = {
 # 代码级回退默认值（实际配置优先从 DB 读取）
 _FALLBACK = {
     "opencode-go": {"base_url": "https://opencode.ai/zen/go/v1", "model": "deepseek-v4-flash"},
-    "nvidia": {"base_url": "https://integrate.api.nvidia.com/v1", "model": "stepfun-ai/step-3.5-flash"},
+    "nvidia": {"base_url": "https://integrate.api.nvidia.com/v1", "model": "stepfun-ai/step-3.7-flash"},
     "sensenova": {"base_url": "https://token.sensenova.cn/v1", "model": "deepseek-v4-flash"},
 }