Implement Phase 12 feedback, Phase 13 Semantic Chunker, Phase 13-B Reranker, Bug 5 thinking fix

- Phase 12: FeedbackRepository + td_feedback 테이블, Gradio 👍/👎 이벤트, run_id 추적, LangSmith create_feedback() 연동
- Phase 13: 커스텀 _SemanticSplitter 제거 → langchain_experimental.SemanticChunker 교체, buffer_size/threshold_type 환경변수 적용
- Phase 13-B: RerankService (Cross-Encoder), RetrieverService.search()에 reranker 통합, tools.py as_retriever() → search() 전환
- Bug 5: mlx_chat_model enable_thinking 런타임 오버라이드, agent_service stream_mode=["messages","custom"] 이중 스트림, thinking 토큰 custom 이벤트로 emit
- ROADMAP: LLM 모델명 8B 반영, RAG에 Reranker 추가, 추천 진행 순서 갱신

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app.py

@@ -1,4 +1,4 @@
-"""Gradio Web UI — 율봇 Phase 4 + Phase 9/10 + Phase 14(음성)."""
+"""Gradio Web UI — 율봇 Phase 4 + Phase 9/10 + Phase 12(피드백) + Phase 14(음성)."""
 import os
 import subprocess
 import tempfile
@@ -17,6 +17,7 @@ db.init_schema()
 
 ingestion = container.ingestion_service()
 retriever = container.retriever_service()
+feedback_repo = container.feedback_repository()
 
 _cfg = container.config()
 _agent_cache: dict[str, AgentService] = {}
@@ -44,7 +45,7 @@ def transcribe_audio(filepath: str) -> str:
 
 
 def tts_speak(text: str, voice: str) -> str | None:
-    """텍스트를 macOS say 명령어로 음성 변환, 재생용 wav 파일 경로 반환."""
+    """텍스트를 macOS say 명령어로 음성 변환, 재생용 aiff 파일 경로 반환."""
     if not text:
         return None
     try:
@@ -77,36 +78,72 @@ def _get_agent(user_id: str) -> AgentService:
     return _agent_cache[user_id]
 
 
-async def respond(message, history, show_thinking, user_id, use_tts):
+async def respond(message, history, show_thinking, user_id, use_tts, run_ids):
     if not message.strip():
-        yield history, "", None
+        yield history, "", None, run_ids
         return
 
     agent = _get_agent(user_id)
     history = list(history)
+    run_ids = list(run_ids)
     history.append({"role": "user", "content": message})
     history.append({"role": "assistant", "content": ""})
-    yield history, "", None
+    yield history, "", None, run_ids
 
     async for token in agent.stream_response(message, show_thinking=show_thinking):
         history[-1]["content"] += token
-        yield history, "", None
+        yield history, "", None, run_ids
+
+    run_ids.append(agent.last_run_id)
 
     if use_tts:
         response_text = history[-1]["content"]
         audio_path = tts_speak(response_text, _cfg.tts_voice)
-        yield history, "", audio_path
+        yield history, "", audio_path, run_ids
+    else:
+        yield history, "", None, run_ids
+
+
+def handle_feedback(like_data: gr.LikeData, history, run_ids, user_id):
+    idx = like_data.index
+    if isinstance(idx, (list, tuple)):
+        idx = idx[0]
+    if not isinstance(idx, int) or idx >= len(history):
+        return
+    if history[idx].get("role") != "assistant":
+        return
+    asst_turn = sum(1 for m in history[:idx] if m.get("role") == "assistant")
+    run_id = run_ids[asst_turn] if asst_turn < len(run_ids) else None
+
+    def _to_str(val) -> str:
+        return val if isinstance(val, str) else str(val)
+
+    user_msg = _to_str(history[idx - 1]["content"]) if idx > 0 else ""
+    asst_msg = _to_str(history[idx]["content"])
+    rating = 1 if like_data.liked else -1
+
+    try:
+        feedback_repo.save_feedback(user_id, user_msg, asst_msg, rating, run_id)
+    except Exception as e:
+        print(f"[Feedback] DB 저장 실패: {e}")
+
+    if run_id and os.getenv("LANGCHAIN_TRACING_V2") == "true":
+        try:
+            from langsmith import Client
+            Client().create_feedback(run_id=run_id, key="user_feedback", score=rating)
+        except Exception as e:
+            print(f"[Feedback] LangSmith 기록 실패: {e}")
 
 
 def switch_user(user_id):
-    """사용자 전환 시 채팅 화면만 초기화 (대화 이력은 유지)."""
-    return []
+    """사용자 전환 시 채팅 화면과 run_ids 초기화 (대화 이력은 DB에 유지)."""
+    return [], []
 
 
 def reset_chat(user_id):
     agent = _get_agent(user_id)
     agent.reset()
-    return []
+    return [], []
 
 
 def ingest_files(files):
@@ -143,6 +180,7 @@ with gr.Blocks(title="율봇") as demo:
     gr.Markdown("# 율봇\n육아·금융 전문 AI 상담 도우미")
 
     user_state = gr.State(DEFAULT_USER)
+    run_ids_state = gr.State([])
 
     with gr.Tab("대화"):
         with gr.Row():
@@ -185,7 +223,7 @@ with gr.Blocks(title="율봇") as demo:
         user_selector.change(
             switch_user,
             inputs=[user_selector],
-            outputs=[chatbot],
+            outputs=[chatbot, run_ids_state],
         ).then(
             lambda u: u, inputs=[user_selector], outputs=[user_state]
         )
@@ -198,15 +236,21 @@ with gr.Blocks(title="율봇") as demo:
 
         send_btn.click(
             respond,
-            inputs=[msg_box, chatbot, show_thinking, user_state, use_tts],
-            outputs=[chatbot, msg_box, tts_output],
+            inputs=[msg_box, chatbot, show_thinking, user_state, use_tts, run_ids_state],
+            outputs=[chatbot, msg_box, tts_output, run_ids_state],
         )
         msg_box.submit(
             respond,
-            inputs=[msg_box, chatbot, show_thinking, user_state, use_tts],
-            outputs=[chatbot, msg_box, tts_output],
+            inputs=[msg_box, chatbot, show_thinking, user_state, use_tts, run_ids_state],
+            outputs=[chatbot, msg_box, tts_output, run_ids_state],
+        )
+        reset_btn.click(reset_chat, inputs=[user_state], outputs=[chatbot, run_ids_state])
+
+        chatbot.like(
+            handle_feedback,
+            inputs=[chatbot, run_ids_state, user_state],
+            outputs=[],
         )
-        reset_btn.click(reset_chat, inputs=[user_state], outputs=[chatbot])
 
     with gr.Tab("문서 등록"):
         gr.Markdown("PDF 또는 TXT 파일을 업로드하면 율봇이 내용을 참고해 답변합니다.")