feat: optimize all 13 prompts by function+scenario; fix topic_selector hardcode

Prompt optimizations:
- topics_trends: 10→8 topics, simplified output, anti-hallucination
- topic_generate: stronger structure, user-pain focus
- topic_selector_gaps: add SEO+social instruction, clean JSON output
- section_expansion: shorter (removed over-detailed AI套话 list), cleaner
- title_optimize_*: 3 titles per line (no numbering), cleaned up
- outline_generation: simplified structure, strong anti-generic-titles
- compliance_fix/polish: expanded fix types, stronger output requirements
- sources_optimization: cleaner JSON schema
- tags_generation: added 'no thinking output' instruction
- research_summary: simplified output structure
- clean_ai_verbosity/clean_thinking_patterns: unchanged (rule-based)

topic_selector.py: hardcoded prompt → get_prompt('topic_selector_gaps')
_PROMPT_DEFAULTS seeds updated to match DB

scripts/topic_selector.py

@@ -15,6 +15,7 @@ sys.path.insert(0, str(PROJECT_ROOT / "platform" / "backend"))
 from trends import load_trends
 from db_helper import export_topics_to_json, update_topic_status, save_topics_to_db
 from app.core.nvidia_client import call_llm
+from prompt_loader import get_prompt
 
 DATA_DIR = PROJECT_ROOT / "automation" / "data"
 TRENDS_FILE = DATA_DIR / "trends.json"
@@ -130,23 +131,11 @@ def detect_trend_gaps() -> List[Dict]:
 def generate_new_topics(gaps: List[Dict]) -> List[Dict]:
     if not gaps:
         return []
-    prompt = f"""你是一个敏锐的内容策略师，擅长将热点转化为有价值、有传播力的选题。以下热点当前未覆盖，请为每个热点生成选题建议。
-
-注意：选题要兼具SEO价值（能在搜索中被找到）和社交传播力（能在平台引发讨论）。
-
-热点列表：
-{chr(10).join(f'- {g["topic"]}（{g.get("domain","")}）— {g.get("reason","")}。热搜词：{", ".join(g.get("hot_keywords", []))}）' for g in gaps)}
-
-每个选题需包含：
-- "title": 标题（20字内，包含核心关键词，有吸引力）
-- "field": 所属领域
-- "core_concept": 核心观点（一句话说清独特价值）
-- "audience_pain": 受众痛点（真实用户的困惑/焦虑/需求）
-- "unique_angle": 独特视角（差异化切入点，含SEO关键词潜力）
-- "target_platform": 最适合发布平台（知乎/小红书/微信/多平台）
-- "estimated_search_volume": 预估搜索热度（高/中/低）
-
-只输出 JSON 数组，不要其他文字。"""
+    gaps_text = chr(10).join(
+        f'- {g["topic"]}（{g.get("domain","")}）— {g.get("reason","")}。热搜词：{", ".join(g.get("hot_keywords", []))}）'
+        for g in gaps
+    )
+    prompt = get_prompt("topic_selector_gaps", gaps=gaps_text)
     try:
         resp = call_llm(prompt, temperature=0.4)
         resp = resp.strip()