feat(bootstrap): refactor runtime dependency management and add lazy loading for binary store and vector index

feat(agent): update import for agent session service
feat(openai): add context truncation check in OpenAI answer generator
docs(README): update frontend environment file conventions
fix(vite): default local frontend development to local backend

backend/app/api/main.py

@@ -11,7 +11,7 @@ from app.api.models import ErrorResponse
 from app.api.routes import api_router
 from app.config.logging import setup_logging
 from app.config.settings import settings
-from app.services.llm.llm_factory import LLMFactory
+from app.shared.bootstrap import cleanup_runtime_dependencies, preload_runtime_dependencies
 # Keep module behavior explicit so the backend flow stays easy to audit.
 
 
@@ -24,12 +24,12 @@ async def lifespan(app: FastAPI):
     logger.info(f"启动 {settings.app_name} v{settings.app_version}")
     logger.info(f"调试模式: {settings.debug}")
     logger.info("预加载LLM客户端...")
-    LLMFactory.preload_clients(["qwen", "deepseek"])
+    preload_runtime_dependencies()
 
     yield
 
     logger.info("应用关闭，执行清理...")
-    LLMFactory.cleanup()
+    cleanup_runtime_dependencies()
 
 
 app = FastAPI(

backend/app/api/routes/agent.py

@@ -20,7 +20,7 @@ from app.api.models import (
 )
 from app.config.settings import settings
 from app.shared.async_utils import iter_in_thread
-from app.shared.bootstrap import get_agent_conversation_service, get_conversation_store
+from app.shared.bootstrap import get_agent_conversation_service, get_agent_session_service
 # Keep route handlers close to their transport-layer wiring for easier auditing.