feat(eval): Playground GLM 评分落库并可恢复

在 conversations 表增加 playground_conversation_judge_json，流式/非流式对话评审结束后写入最近一次快照（整体分、逐轮分、对比文案、错误与基线文件名等）。新增只读 GET 供前端按会话拉取；评测台 Playground 切换会话时自动恢复，并提示基线是否和当时一致。

api/app/features/evaluation/schemas.py

@@ -211,6 +211,13 @@ class ManualJudgeConversationOut(BaseModel):
     errors: list[str] = Field(default_factory=list)
 
 
+class PlaygroundConversationJudgeOut(BaseModel):
+    """`conversations.playground_conversation_judge_json` 的只读视图。"""
+
+    conversation_id: str
+    judge: dict[str, Any] | None = None
+
+
 class ManualJudgeMemoirBody(BaseModel):
     user_id: str
     baseline_sections: list[MemoirSectionBaselineOut] | None = None