refactor(api): TOML 配置 SSOT、统一错误契约、Auth/事务加固与可观测性 (#33)

配置 SSOT（TOML + .env）
统一错误契约
Auth 与事务边界
Redis / Celery 可靠性:业务 Redis（DB/0）与 Celery broker/backend（DB/1）显式拆分；连接池、sync client
可观测性（OpenTelemetry + LGTM）

api/app/agents/chat/interview_agent.py

@@ -48,6 +48,9 @@ from app.core.config import settings
 from app.core.llm_gateway import LlmGateway, LlmUseCase
 from app.core.logging import get_logger
 from app.features.conversation.input_normalize import normalize_chat_input_for_agent
+from app.core.runtime_constants import agent_log_defaults
+from app.features.conversation.constants import chat
+from app.features.story.constants import story
 
 logger = get_logger(__name__)
 
@@ -171,8 +174,8 @@ class InterviewAgent:
         if normalized_user_message is not None:
             return (normalized_user_message or "").strip()
         llm_n = None
-        if settings.chat_input_normalize_enabled and (
-            (settings.chat_input_normalize_mode or "").strip().lower() == "llm"
+        if chat.input_normalize_enabled and (
+            (chat.input_normalize_mode or "").strip().lower() == "llm"
         ):
             llm_n = self.llm
         return normalize_chat_input_for_agent(user_message or "", llm=llm_n)
@@ -218,16 +221,16 @@ class InterviewAgent:
                 du = self._detect_user_stage(text_for_model)
             hw = await get_history_with_window(
                 conversation_id,
-                max_pairs=settings.chat_history_max_pairs,
-                max_chars=settings.chat_history_max_chars,
+                max_pairs=chat.history_max_pairs,
+                max_chars=chat.history_max_chars,
             )
             recent_questions = extract_recent_questions(hw.window)
             conversation_turn_total = hw.turn_total
             all_stages_coverage = narrative_state.all_stages_coverage()
-            persona = normalize_interview_persona(settings.chat_interview_persona)
-            max_segments = int(settings.chat_interview_max_segments)
-            max_tokens = int(settings.chat_interview_max_tokens)
-            max_chars = int(settings.chat_interview_max_chars_per_segment)
+            persona = normalize_interview_persona(chat.interview_persona)
+            max_segments = int(chat.interview_max_segments)
+            max_tokens = int(chat.interview_max_tokens)
+            max_chars = int(chat.interview_max_chars_per_segment)
 
             turn_plan = plan_interview_turn(
                 current_stage=memoir_state.current_stage,
@@ -246,7 +249,7 @@ class InterviewAgent:
             reply_planner_raw = ""
             baseline_mode = turn_plan.mode
             baseline_primary_focus = turn_plan.primary_focus
-            if settings.chat_reply_planner_llm_enabled:
+            if chat.reply_planner_llm_enabled:
                 rq_preview = (
                     "\n".join(recent_questions[-4:])
                     if recent_questions
@@ -258,8 +261,8 @@ class InterviewAgent:
                     text_for_model=text_for_model,
                     memory_evidence_text=(memory_planner_text or memory_evidence_text)
                     or "",
-                    max_tokens=int(settings.chat_reply_planner_max_tokens),
-                    temperature=float(settings.chat_reply_planner_temperature),
+                    max_tokens=int(chat.reply_planner_max_tokens),
+                    temperature=float(chat.reply_planner_temperature),
                     scene_cues_for_planner=scene_cues_for_planner or [],
                     recent_questions_preview=rq_preview,
                 )
@@ -310,12 +313,12 @@ class InterviewAgent:
                 "InterviewAgent.generate_response.prompt",
                 format_history_string(
                     messages,
-                    omit_system_body=settings.agent_log_omit_system_message_body,
+                    omit_system_body=agent_log_defaults.omit_system_message_body,
                 ),
             )
             chat_llm = self.llm.bind(
                 max_tokens=max_tokens,
-                temperature=float(settings.chat_interview_temperature),
+                temperature=float(chat.interview_temperature),
             )
             prompt_chars = _message_contents_char_count(messages)
             llm_t0 = time.perf_counter()
@@ -377,7 +380,7 @@ class InterviewAgent:
                     "InterviewAgent.generate_response.retry_prompt",
                     format_history_string(
                         retry_messages,
-                        omit_system_body=settings.agent_log_omit_system_message_body,
+                        omit_system_body=agent_log_defaults.omit_system_message_body,
                     ),
                 )
                 llm_t1 = time.perf_counter()
@@ -445,7 +448,7 @@ class InterviewAgent:
                     "duplicate_question_guard_llm_retry": retry_used,
                     "autobiographical_boundary_guard_triggered": auto_bio,
                     "reply_planner_llm_used": bool(
-                        settings.chat_reply_planner_llm_enabled
+                        chat.reply_planner_llm_enabled
                         and (reply_planner_raw or "").strip()
                     ),
                     "reply_planner_raw_preview": (reply_planner_raw or "")[:800],
@@ -483,7 +486,7 @@ class InterviewAgent:
             )
             slot_table = SLOT_NAME_MAP_EN if language == "en" else SLOT_NAME_MAP
             empty_slots_readable = [slot_table.get(s, s) for s in empty_slots]
-            persona = normalize_interview_persona(settings.chat_interview_persona)
+            persona = normalize_interview_persona(chat.interview_persona)
             prompt = get_opening_prompt(
                 current_stage=memoir_state.current_stage,
                 empty_slots_readable=empty_slots_readable,
@@ -497,8 +500,8 @@ class InterviewAgent:
             )
             hw = await get_history_with_window(
                 conversation_id,
-                max_pairs=settings.chat_history_max_pairs,
-                max_chars=settings.chat_history_max_chars,
+                max_pairs=chat.history_max_pairs,
+                max_chars=chat.history_max_chars,
             )
             messages: List[Any] = [SystemMessage(content=prompt)]
             messages.extend(hw.window)
@@ -520,12 +523,12 @@ class InterviewAgent:
                 "InterviewAgent.opening.prompt",
                 format_history_string(
                     messages,
-                    omit_system_body=settings.agent_log_omit_system_message_body,
+                    omit_system_body=agent_log_defaults.omit_system_message_body,
                 ),
             )
             opening_llm = self.llm.bind(
-                max_tokens=settings.chat_opening_max_tokens,
-                temperature=float(settings.chat_interview_temperature),
+                max_tokens=chat.opening_max_tokens,
+                temperature=float(chat.interview_temperature),
             )
             prompt_chars = _message_contents_char_count(messages)
             llm_t0 = time.perf_counter()
@@ -564,7 +567,7 @@ class InterviewAgent:
             raw_list = segments_from_llm_response(response_text, max_segments=2)
             if not raw_list:
                 raw_list = [response_text.strip()]
-            max_chars = int(settings.chat_interview_max_chars_per_segment)
+            max_chars = int(chat.interview_max_chars_per_segment)
             out = truncate_chat_segments(
                 raw_list,
                 max_segments=2,
@@ -612,7 +615,7 @@ class InterviewAgent:
             )
             slot_table = SLOT_NAME_MAP_EN if language == "en" else SLOT_NAME_MAP
             empty_slots_readable = [slot_table.get(s, s) for s in empty_slots]
-            persona = normalize_interview_persona(settings.chat_interview_persona)
+            persona = normalize_interview_persona(chat.interview_persona)
             prompt = get_re_greeting_prompt(
                 current_stage=memoir_state.current_stage,
                 empty_slots_readable=empty_slots_readable,
@@ -627,8 +630,8 @@ class InterviewAgent:
             )
             hw = await get_history_with_window(
                 conversation_id,
-                max_pairs=settings.chat_history_max_pairs,
-                max_chars=settings.chat_history_max_chars,
+                max_pairs=chat.history_max_pairs,
+                max_chars=chat.history_max_chars,
             )
             messages: List[Any] = [SystemMessage(content=prompt)]
             messages.extend(hw.window)
@@ -647,12 +650,12 @@ class InterviewAgent:
                 "InterviewAgent.re_greeting.prompt",
                 format_history_string(
                     messages,
-                    omit_system_body=settings.agent_log_omit_system_message_body,
+                    omit_system_body=agent_log_defaults.omit_system_message_body,
                 ),
             )
             re_greet_llm = self.llm.bind(
-                max_tokens=settings.chat_opening_max_tokens,
-                temperature=float(settings.chat_interview_temperature),
+                max_tokens=chat.opening_max_tokens,
+                temperature=float(chat.interview_temperature),
             )
             llm_t0 = time.perf_counter()
             with agent_span(
@@ -691,7 +694,7 @@ class InterviewAgent:
             raw_list = segments_from_llm_response(response_text, max_segments=2)
             if not raw_list:
                 raw_list = [response_text.strip()]
-            max_chars = int(settings.chat_interview_max_chars_per_segment)
+            max_chars = int(chat.interview_max_chars_per_segment)
             out = truncate_chat_segments(
                 raw_list,
                 max_segments=2,