refactor(memory): restructure memory system and improve configuration management

- Remove deprecated main.py entry point from memory module
- Reorganize imports across controllers and services for consistency
- Update emotion controller to pass db session instead of config_id to services
- Enhance memory agent controller with db session parameter for status_type and user_profile endpoints
- Refactor memory agent service to accept db parameter in classify_message_type method
- Improve configuration handling in celery_app by removing automatic database reload
- Update all memory-related services to use centralized config management
- Standardize import ordering and remove unused imports across 50+ files
- Add pilot_run_service for new pilot execution workflow
- Refactor extraction engine, reflection engine, and search services for better modularity
- Update LLM utilities and embedder configuration for improved flexibility
- Enhance type classifier and verification tools with better error handling
- Improve memory evaluation modules (LOCOMO, LongMemEval, MemSciQA) with consistent patterns

api/app/core/rag_utils/chunk_insight.py

@@ -5,16 +5,24 @@ This module provides functionality to analyze chunk content and generate insight
 """
 
 import asyncio
-from typing import List, Dict, Any
 from collections import Counter
+from typing import Any, Dict, List
+
+from app.core.logging_config import get_business_logger
+from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
+from app.db import get_db_context
 from pydantic import BaseModel, Field
 
-from app.core.memory.utils.llm.llm_utils import get_llm_client
-from app.core.logging_config import get_business_logger
-
 business_logger = get_business_logger()
 
 
+def _get_llm_client():
+    """Get LLM client using db context."""
+    with get_db_context() as db:
+        factory = MemoryClientFactory(db)
+        return factory.get_llm_client(None)  # Uses default LLM
+
+
 class ChunkInsight(BaseModel):
     """Pydantic model for chunk insight."""
     insight: str = Field(..., description="对chunk内容的深度洞察分析")
@@ -40,7 +48,7 @@ async def classify_chunk_domain(chunk: str) -> str:
         Domain name
     """
     try:
-        llm_client = get_llm_client()
+        llm_client = _get_llm_client()
         
         prompt = f"""请将以下文本内容归类到最合适的领域中。
 
@@ -177,7 +185,7 @@ async def generate_chunk_insight(chunks: List[str], max_chunks: int = 15) -> str
         ]
         
         # 调用LLM生成洞察
-        llm_client = get_llm_client()
+        llm_client = _get_llm_client()
         response = await llm_client.chat(messages=messages)
         
         insight = response.content.strip()

api/app/core/rag_utils/chunk_summary.py

@@ -5,15 +5,23 @@ This module provides functionality to summarize chunk content using LLM.
 """
 
 import asyncio
-from typing import List, Dict, Any
+from typing import Any, Dict, List
+
+from app.core.logging_config import get_business_logger
+from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
+from app.db import get_db_context
 from pydantic import BaseModel, Field
 
-from app.core.memory.utils.llm.llm_utils import get_llm_client
-from app.core.logging_config import get_business_logger
-
 business_logger = get_business_logger()
 
 
+def _get_llm_client():
+    """Get LLM client using db context."""
+    with get_db_context() as db:
+        factory = MemoryClientFactory(db)
+        return factory.get_llm_client(None)  # Uses default LLM
+
+
 class ChunkSummary(BaseModel):
     """Pydantic model for chunk summary."""
     summary: str = Field(..., description="简洁的chunk内容摘要")
@@ -59,7 +67,7 @@ async def generate_chunk_summary(chunks: List[str], max_chunks: int = 10) -> str
         ]
         
         # 调用LLM生成摘要
-        llm_client = get_llm_client()
+        llm_client = _get_llm_client()
         response = await llm_client.chat(messages=messages)
         
         summary = response.content.strip()

api/app/core/rag_utils/chunk_tags.py

@@ -7,14 +7,22 @@ This module provides functionality to extract meaningful tags from chunk content
 import asyncio
 from collections import Counter
 from typing import List, Tuple
+
+from app.core.logging_config import get_business_logger
+from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
+from app.db import get_db_context
 from pydantic import BaseModel, Field
 
-from app.core.memory.utils.llm.llm_utils import get_llm_client
-from app.core.logging_config import get_business_logger
-
 business_logger = get_business_logger()
 
 
+def _get_llm_client():
+    """Get LLM client using db context."""
+    with get_db_context() as db:
+        factory = MemoryClientFactory(db)
+        return factory.get_llm_client(None)  # Uses default LLM
+
+
 class ExtractedTags(BaseModel):
     """Pydantic model for extracted tags."""
     tags: List[str] = Field(..., description="从文本中提取的关键标签列表")
@@ -56,7 +64,7 @@ async def extract_chunk_tags(chunks: List[str], max_tags: int = 10, max_chunks:
             "标签应该是名词或名词短语，能够准确概括文本的核心内容。"
         )
         
-        llm_client = get_llm_client()
+        llm_client = _get_llm_client()
         
         # 为每个chunk单独提取标签，然后统计频率
         all_tags = []
@@ -151,7 +159,7 @@ async def extract_chunk_persona(chunks: List[str], max_personas: int = 5, max_ch
         ]
         
         # 调用LLM提取人物形象
-        llm_client = get_llm_client()
+        llm_client = _get_llm_client()
         structured_response = await llm_client.response_structured(
             messages=messages,
             response_model=ExtractedPersona