fix(lead): Ensure Strict AI Selection via product_ids, Prevent search fallbacks and add proper Device ID logic

- Classifier (1 LLM call): classify stage via STAGE_JSON + call tools
- Stylist (1 LLM call): format response with optional stage context
- Removed _TOOL_STAGE_MAP hardcoded heuristic (stage is optional)
- Added re import for STAGE_JSON regex parsing
- Cleaned up debug logging

- New diagram agent: Planner generates Mermaid code, Responder explains
- FastAPI route: /api/diagram/chat + /clear with Redis session history
- Frontend: split-pane UI with canvas pan/zoom (scroll wheel + drag)
- Auto-fit diagram to view, grid background, PNG export
- Prompt enforces ASCII syntax labels for Mermaid compatibility
- Planner LLM max_tokens=4000 for complex diagram generation
- Supports: flowchart, sequence, class, ER, gantt, mindmap, pie chart

- HITL: Accept/Reject SQL before execution
- Chart.js: Auto-detect data shape -> render number cards/bar/doughnut/line
- Prompt: Verified StarRocks partition pruning rules (no DATE() on column)
- Summary: Agent provides 2-3 sentence Vietnamese explanation of results
- Fix: renderTraceDocument -> renderDataResults for history loading
- Auth: Login/admin pages, middleware, JWT auth
- Report Agent: Full report generation system
- New pages: stress-test, regression-test, user-simulator, competitor-research

- Add AI analytic dashboard (ai-analytic.html)
- Add report agent route and prompt
- Add langfuse analytics module
- Update middleware and server config

feat: add chat history analytics, product guide, experiment edit, docker network 10.101.10.x, workers=1