feat: AI analysis engine refactor, dark theme polish & virtual position management

Core changes:
- Refactor FastAnalysisService: single LLM multi-factor analysis replaces
  7-agent pipeline; add multi-timeframe consensus, threshold calibration,
  confidence calibration, multi-model ensemble voting
- Add RAG memory injection and reflection validation (analysis_memory +
  reflection worker)
- Simplify billing config: remove unused strategy_run/backtest/portfolio_monitor,
  add ai_code_gen separate billing (different token consumption scale)
- Settings hot-reload after save, no backend restart needed

Frontend:
- Global dark theme overhaul: pure black palette replacing blue-tinted colors
  across sidebar/header/dashboard/analysis/K-line/user-manage/profile/settings/billing
- Fix USDT payment modal dark theme (portal rendering broke CSS selectors)
- Refactor position modal: direction + quantity + entry price, remove add/reduce
  logic, show raw DB values on re-open, save exactly what user inputs
- Fix Polymarket prediction market dark text
- i18n for position modal title

Backend:
- Position management: one record per symbol (DELETE+INSERT replacing
  ON CONFLICT with side), fixes PnL showing 0 when switching long/short
- MarketDataCollector data fetching optimization
- portfolio_monitor scheduled monitoring improvements
- env.example reorganized: common config first, advanced config last

Documentation:
- README architecture diagram updated to FastAnalysisService flow
- Add virtual position, AI tuning config, billing items documentation
- Add INDICATOR_DEFINITIONS_CN.md, FRONTEND_FAST_ANALYSIS.md

Made-with: Cursor

backend_api_python/app/routes/indicator.py

@@ -698,7 +698,18 @@ IMPORTANT: Output Python code directly, without explanations, without descriptio
         return content.strip() or _template_code()
 
     def stream():
-        # 不扣任何 QDT：开源本地版直接生成/返回代码
+        from app.services.billing_service import get_billing_service
+        billing = get_billing_service()
+        ok, msg = billing.check_and_consume(
+            user_id=g.user_id,
+            feature='ai_code_gen',
+            reference_id=f"ai_code_gen_{g.user_id}_{int(time.time())}"
+        )
+        if not ok:
+            yield "data: " + json.dumps({"error": f"积分不足: {msg}"}, ensure_ascii=False) + "\n\n"
+            yield "data: [DONE]\n\n"
+            return
+
         try:
             code_text = _generate_code_via_llm()
         except Exception as e: