feat: enhance caching mechanisms to prevent memory leaks

- Improved in-memory rate limiting by evicting timestamps outside the current window and cleaning up empty keys.
- Updated LLM router service to cache context profiles and avoid redundant computations.
- Introduced cache eviction logic for MCP tools and sandbox instances to manage memory usage effectively.
- Added garbage collection triggers in chat streaming functions to reclaim resources promptly.

surfsense_backend/app/agents/new_chat/llm_config.py

@@ -22,6 +22,7 @@ from app.services.llm_router_service import (
     AUTO_MODE_ID,
     ChatLiteLLMRouter,
     LLMRouterService,
+    get_auto_mode_llm,
     is_auto_mode,
 )
 
@@ -389,7 +390,7 @@ def create_chat_litellm_from_agent_config(
             print("Error: Auto mode requested but LLM Router not initialized")
             return None
         try:
-            return ChatLiteLLMRouter()
+            return get_auto_mode_llm()
         except Exception as e:
             print(f"Error creating ChatLiteLLMRouter: {e}")
             return None