perf(kb-planner): route internal planner calls to dedicated small/fast LLM

Adds an optional planner LLM role wired through KnowledgePriorityMiddleware
so KB query rewriting, date extraction, and recency classification run on a
cheap model (e.g. gpt-4o-mini, Haiku, Azure nano) instead of the user's
chat LLM. Operators opt in by setting is_planner: true on exactly one
global config; without it, behavior is unchanged.

surfsense_backend/app/agents/multi_agent_chat/middleware/main_agent/knowledge_priority.py

@@ -6,6 +6,7 @@ from langchain_core.language_models import BaseChatModel
 
 from app.agents.new_chat.filesystem_selection import FilesystemMode
 from app.agents.new_chat.middleware import KnowledgePriorityMiddleware
+from app.services.llm_service import get_planner_llm
 
 
 def build_knowledge_priority_mw(
@@ -19,6 +20,7 @@ def build_knowledge_priority_mw(
 ) -> KnowledgePriorityMiddleware:
     return KnowledgePriorityMiddleware(
         llm=llm,
+        planner_llm=get_planner_llm(),
         search_space_id=search_space_id,
         filesystem_mode=filesystem_mode,
         available_connectors=available_connectors,