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/frontend/admin.html

@@ -608,7 +608,7 @@
                     <el-col :span="12"><el-form-item label="供应商">
                         <el-select v-model="llmConfigForm.provider" style="width:100%">
                             <el-option label="opencode-go (deepseek-v4-flash)" value="opencode-go"></el-option>
-                            <el-option label="nvidia (step-3.5-flash)" value="nvidia"></el-option>
+                            <el-option label="nvidia (step-3.7-flash)" value="nvidia"></el-option>
                             <el-option label="sensenova (deepseek-v4-flash)" value="sensenova"></el-option>
                         </el-select>
                     </el-form-item></el-col>