refactor(backend): move _BOUNDARY_SCAN_LIMIT to module level, logger.info -> debug

Move _BOUNDARY_SCAN_LIMIT from function body to module level to avoid
redefining it on every backward-pagination call. Downgrade GET_SESSION
per-request log from INFO to DEBUG to reduce noise at scale.

autogpt_platform/backend/backend/api/features/chat/routes.py

@@ -543,7 +543,7 @@ async def get_session(
         _strip_injected_context(message.model_dump()) for message in page.messages
     ]
 
-    logger.info(
+    logger.debug(
         "[GET_SESSION] session=%s, active=%s, from_start=%s, forward_paginated=%s, "
         "msg_count=%d, last_role=%s",
         session_id,

autogpt_platform/backend/backend/copilot/db.py

@@ -30,6 +30,8 @@ from .model import get_chat_session as get_chat_session_cached
 
 logger = logging.getLogger(__name__)
 
+_BOUNDARY_SCAN_LIMIT = 10
+
 
 class PaginatedMessages(BaseModel):
     """Result of a paginated message query."""
@@ -124,7 +126,6 @@ async def get_chat_messages_paginated(
         # expand backward to include the preceding assistant message that
         # owns the tool_calls, so convertChatSessionMessagesToUiMessages
         # can pair them correctly.
-        _BOUNDARY_SCAN_LIMIT = 10
         if results and results[0].role == "tool":
             boundary_where: dict[str, Any] = {
                 "sessionId": session_id,