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/features/memoir/story_pipeline_sync.py

@@ -27,7 +27,6 @@ from app.agents.memoir.story_route_agent import (
     StoryRouteAgent,
     default_append_target_story_id,
 )
-from app.core.business_telemetry import business_span
 from app.agents.stage_constants import (
     CATEGORY_TO_CHAT_STAGE,
     CHAPTER_CATEGORIES,
@@ -35,6 +34,7 @@ from app.agents.stage_constants import (
     STAGE_TO_ORDER,
 )
 from app.agents.state_schema import MemoirStateSchema
+from app.core.business_telemetry import business_span
 from app.core.config import settings
 from app.core.logging import get_logger
 from app.features.conversation.lineage_schemas import aggregate_lineage_from_segments
@@ -61,6 +61,8 @@ from app.features.memoir.repo import (
 )
 from app.features.memory.evidence_format import format_evidence_chunks_for_prompt
 from app.features.story.models import Story, StoryVersion
+from app.features.memoir.constants import memoir
+from app.features.story.constants import story
 from app.features.story.sync_write import (
     append_story_version_sync,
     count_story_versions_sync,
@@ -267,7 +269,7 @@ def _title_slots_filtered_for_generation(
     slot_snippets: dict[str, str], *, md: str, oral_scope: str
 ) -> dict[str, str]:
     """仅保留与正文或本批口述有文本重叠的 slot，降低档案/历史 slot 串台到标题。"""
-    if not settings.memoir_title_slots_require_body_or_oral_match:
+    if not memoir.title_slots_require_body_or_oral_match:
         return dict(slot_snippets)
     hay = f"{(md or '').strip()}\n{(oral_scope or '').strip()}"
     if not hay.strip():
@@ -311,7 +313,7 @@ def _strip_ungrounded_title_segments(
     """
     按 · / • 分节丢弃含未落地履历短语的小节；全部丢弃则占位。
     """
-    if not settings.memoir_title_hay_grounding_strict_phrases_enabled:
+    if not memoir.title_hay_grounding_strict_phrases_enabled:
         return (title or "").strip() or _placeholder_title(
             chapter_category, language=language
         )
@@ -358,7 +360,7 @@ def _maybe_generate_title(
 ) -> str:
     """Generate a title only when body is long enough; otherwise return placeholder."""
     body_len = len((md or "").strip())
-    if body_len < settings.story_title_min_body_chars:
+    if body_len < story.title_min_body_chars:
         return _placeholder_title(chapter_category, language=language)
     content_excerpt = (md or "").strip()[:300]
     merged_slots = _title_slots_filtered_for_generation(
@@ -392,8 +394,8 @@ def _route_segment_texts(category_segments: list) -> list[tuple[str, str]]:
     for seg in category_segments:
         raw = seg.user_input_text or ""
         if (
-            settings.memoir_oral_normalize_enabled
-            and (settings.memoir_oral_normalize_mode or "rules").strip().lower()
+            memoir.oral_normalize_enabled
+            and (memoir.oral_normalize_mode or "rules").strip().lower()
             != "off"
         ):
             t = apply_oral_rules(raw)
@@ -435,7 +437,7 @@ def _gate_narrative_fidelity(
     from app.agents.memoir.fidelity_check_agent import FidelityCheckAgent
 
     check_llm = fidelity_llm if fidelity_llm is not None else llm
-    if not settings.memoir_fidelity_check_enabled or not check_llm:
+    if not memoir.fidelity_check_enabled or not check_llm:
         return narrative_raw, "none"
     agent = FidelityCheckAgent()
     ex = (existing_canonical or "").strip() or None
@@ -471,7 +473,7 @@ def _apply_narrative_body_safety(
     m = (md or "").strip()
     ex = (existing_for_narrative or "").strip()
     o = (oral or "").strip()
-    min_len = int(settings.memoir_narrative_evidence_overlap_min_chars)
+    min_len = int(memoir.narrative_evidence_overlap_min_chars)
     ev_plain = strip_evidence_for_overlap_check(evidence_text)
     if m and body_contains_prompt_artifact(m):
         logger.warning(
@@ -494,7 +496,7 @@ def _apply_narrative_body_safety(
             "evidence_leak_heuristic"
         )
     if (
-        settings.memoir_evidence_scene_anchor_check_enabled
+        memoir.evidence_scene_anchor_check_enabled
         and m
         and evidence_text.strip()
         and evidence_scene_anchor_leak(m, ev_plain, o, ex)
@@ -667,8 +669,8 @@ def _resolve_append_target(
     memoir_correlation_id: str | None,
 ) -> tuple[str | None, str, str]:
     """Resolve append target and return (target_story_id, existing_for_narrative, decision_source)."""
-    max_chars = int(settings.story_append_max_canonical_chars)
-    max_ver = int(settings.story_append_max_versions)
+    max_chars = int(story.append_max_canonical_chars)
+    max_ver = int(story.append_max_versions)
     target_story_id: str | None = None
     existing_for_narrative = ""
 
@@ -696,7 +698,7 @@ def _resolve_append_target(
         and candidate_stories
         and decision_source not in FALLBACK_NEW_STORY_REASONS
         and len(oral_norm)
-        <= int(settings.memoir_story_route_append_guardrail_oral_chars)
+        <= int(memoir.story_route_append_guardrail_oral_chars)
     ):
         tid_g = default_append_target_story_id(candidate_stories, story_meta, settings)
         if tid_g:
@@ -817,7 +819,13 @@ def _execute_narrative_unit(
 
     if target_story_id:
         sid_s = str(target_story_id)
-        ver = append_story_version_sync(session, sid_s, md)
+        try:
+            ver = append_story_version_sync(session, sid_s, md)
+        except ValueError as exc:
+            logger.warning(
+                "append_story_version_sync failed story_id={}: {}", sid_s, exc
+            )
+            return None
         _persist_story_lineage_sync(
             session,
             story_id=sid_s,
@@ -1049,9 +1057,9 @@ def _run_story_pipeline_batch_inner(
     source_ids = [seg.id for seg in category_segments]
 
     n_units = len(category_segments)
-    top_k = int(settings.evidence_top_k_default)
-    if n_units > int(settings.evidence_large_batch_threshold):
-        top_k = int(settings.evidence_top_k_large_batch)
+    top_k = int(story.evidence_top_k_default)
+    if n_units > int(story.evidence_large_batch_threshold):
+        top_k = int(story.evidence_top_k_large_batch)
 
     def _oral_job() -> tuple[str, float]:
         with business_span("memoir.story_pipeline.oral_normalize"):
@@ -1178,7 +1186,7 @@ def _run_story_pipeline_batch_inner(
         plan is None
         and single_route is not None
         and single_route.reason in FALLBACK_NEW_STORY_REASONS
-        and bool(settings.memoir_route_defer_enabled)
+        and bool(memoir.route_defer_enabled)
     ):
         defer_ids = [str(s.id) for s in category_segments]
         logger.info(