feat(eval): server-side replay/phase1 timing + memoir phase1 batch chunking

- Replay and memoir-submit responses include started/finished UTC and elapsed_ms;
  Phase1 poll exposes Redis-backed submit time and elapsed_ms_since_submit.
- Phase1 batch LLM splits segments by memoir_phase1_batch_llm_chunk_size with
  bisect fallback per chunk; Playground shows server timings.

Made-with: Cursor

api/app/features/evaluation/schemas.py

@@ -103,12 +103,25 @@ class ReplayConversationOut(BaseModel):
         default_factory=list,
         description="本批请求创建并已走 orchestrator 的用户 segment id（顺序与落库一致）",
     )
+    #: 服务端计量：本 HTTP 请求内回放逻辑耗时（与浏览器轮询间隔无关）
+    started_at_utc: datetime | None = None
+    finished_at_utc: datetime | None = None
+    elapsed_ms: int | None = Field(
+        default=None,
+        description="服务端 wall 耗时（本请求内 replay_utterances / replay_fixture）",
+    )
 
 
 class MemoirPhase1ReadyOut(BaseModel):
     ready: bool
     checked_segment_ids: list[str] = Field(default_factory=list)
     pending_segment_ids: list[str] = Field(default_factory=list)
+    #: 最近一次 Playground memoir-submit 写入 Redis 的提交时间（无记录时为 None）
+    job_submitted_at_utc: datetime | None = None
+    #: 自 job_submitted_at_utc 至本响应生成时服务端经过的毫秒数
+    elapsed_ms_since_submit: int | None = Field(default=None, ge=0)
+    #: 可选分步耗时（毫秒），键由服务端定义
+    durations_ms: dict[str, int] = Field(default_factory=dict)
 
 
 class MemoirSubmitOut(BaseModel):
@@ -116,6 +129,9 @@ class MemoirSubmitOut(BaseModel):
     user_id: str
     segment_ids: list[str] = Field(default_factory=list)
     celery_task_id: str | None = None
+    submitted_at_utc: datetime | None = None
+    #: 提交接口瞬间耗时，通常为 0；与 Phase1 Celery 执行时间无关
+    elapsed_ms: int | None = Field(default=None, ge=0)
 
 
 class ManualJudgeConversationBody(BaseModel):