feat(memory,conversation): 记忆富化/证据包、时间线幂等字段与对话分段全链路

数据库
- 新增迁移 0003：timeline_events.memory_source_id 外键 → memory_sources，便于按 ingest 源做时间线幂等

后端 - 记忆
- 新增 ingest 后 LLM 富化（摘要/事实/时间线），可配置开关与最大字符数
- 新增证据包组装：合并 chunk、摘要、事实、时间线、故事等检索结果；支持空 query 时是否仍带 rolling 等开关
- repo/retriever/service/router/schemas/summarizer/timeline/extractor 等扩展；文档 memory-retrieval.md 更新

后端 - 对话 WS
- 增加 PING/PONG；分段 ASR 日志与空音频处理；转写失败与「无助手回复」错误提示更明确
- 助手多段回复持久化使用统一分隔符，与分段逻辑一致

后端 - Agent
- reply_limits：按 [SPLIT] 与段落拆段，并保证非空 fallback，供 WS 与 TTS 多段下发

后端 - 回忆录任务
- transcript ingest 记录 source_id；任务成功结?

api/app/features/memory/llm_schemas.py

@@ -0,0 +1,103 @@
+"""LLM JSON 输出校验（memory 富化）。"""
+
+from __future__ import annotations
+
+import json
+from typing import Any, TypeVar
+
+from pydantic import BaseModel, Field, field_validator
+
+TModel = TypeVar("TModel", bound=BaseModel)
+
+
+class ExtractedFactItem(BaseModel):
+    fact_type: str = "event"
+    subject: str | None = None
+    predicate: str | None = None
+    object_json: Any = None
+    confidence: float = Field(default=0.75, ge=0.0, le=1.0)
+    source_chunk_id: str | None = None
+
+    @field_validator("fact_type", mode="before")
+    @classmethod
+    def _coerce_fact_type(cls, v: object) -> str:
+        ft = str(v or "event").strip() or "event"
+        if ft not in ("person", "event", "relation", "place", "milestone"):
+            return "event"
+        return ft
+
+
+class FactsExtractionPayload(BaseModel):
+    facts: list[ExtractedFactItem] = Field(default_factory=list)
+
+
+class SessionSummaryPayload(BaseModel):
+    summary: str = ""
+
+
+class RollingSummaryPayload(BaseModel):
+    rolling_summary: str = ""
+
+
+class TimelineEventItem(BaseModel):
+    event_year: int | None = None
+    event_date: str | None = None
+    title: str = ""
+    description: str | None = None
+    source_fact_ids: list[str] = Field(default_factory=list)
+
+    @field_validator("source_fact_ids", mode="before")
+    @classmethod
+    def _coerce_sf(cls, v: object) -> list[str]:
+        if v is None:
+            return []
+        if isinstance(v, str):
+            return [v] if v else []
+        if isinstance(v, list):
+            return [str(x) for x in v if x]
+        return []
+
+
+class TimelineEventsPayload(BaseModel):
+    events: list[TimelineEventItem] = Field(default_factory=list)
+
+
+def parse_json_payload(raw: str, model: type[TModel]) -> TModel | None:
+    """解析 invoke_json_object 返回的 JSON 字符串。"""
+    from app.features.memoir.memoir_images.json_payload import extract_json_payload
+
+    try:
+        cleaned = extract_json_payload(raw)
+        data = json.loads(cleaned)
+        return model.model_validate(data)
+    except (json.JSONDecodeError, ValueError, TypeError):
+        return None
+
+
+def facts_payload_to_dicts(payload: FactsExtractionPayload) -> list[dict]:
+    out: list[dict] = []
+    for item in payload.facts:
+        d = item.model_dump()
+        scid = d.get("source_chunk_id")
+        if scid is not None and not isinstance(scid, str):
+            d["source_chunk_id"] = str(scid)
+        out.append(d)
+    return out
+
+
+def timeline_payload_to_dicts(payload: TimelineEventsPayload) -> list[dict]:
+    out: list[dict] = []
+    for ev in payload.events:
+        title = (ev.title or "").strip()
+        if not title:
+            continue
+        out.append(
+            {
+                "event_year": ev.event_year,
+                "event_date": ev.event_date,
+                "title": title,
+                "description": ev.description,
+                "source_fact_ids": ev.source_fact_ids or [],
+            }
+        )
+    return out[:20]