WIP: memory system improvements (in progress)

Interview/chat prompt layers, reply planner, style profiles, memory
injection, interview meta store, and related tests. Work not finished.

Made-with: Cursor

api/app/core/config.py

@@ -90,6 +90,12 @@ class Settings(BaseSettings):
     chat_memory_retrieval_enabled: bool = True
     chat_memory_top_k: int = Field(default=8, ge=1, le=30)
     chat_memory_evidence_max_chars: int = Field(default=4096, ge=256, le=50_000)
+    # True：访谈记忆注入使用聊天专用安全格式化（编号引用 + 主语弱化说明）；False 回退 legacy 压平文本
+    chat_memory_safe_evidence_format_enabled: bool = True
+    # True：在规则 TurnPlan 之后追加一轮轻量 JSON focus planner（本轮承接重点 + memory 引用 + 回复形状；失败则回退基线）
+    chat_reply_planner_llm_enabled: bool = False
+    chat_reply_planner_max_tokens: int = Field(default=256, ge=64, le=1024)
+    chat_reply_planner_temperature: float = Field(default=0.2, ge=0.0, le=1.0)
 
     # ── Memoir 叙事忠实度检查（FidelityCheckAgent）────────────────
     memoir_fidelity_check_enabled: bool = True