max_token in route_backend_chat

Note: Two summary calls in route_backend_chats.py have the same max_tokens usage, but they're wrapped in try/except and fail silently (chat still works, just without summarization). These only trigger on long conversations with specific admin settings enabled. The main chat call doesn't set any token limit so it works fine with all models.

Start in route_backend_chats.py and inspect the two summary calls that pass max_tokens, then reproduce the long-conversation path with the relevant admin settings. The issue does not define the intended replacement behavior or a test, so confirm those expectations before changing code; done should include coverage for the summary path without affecting normal chat.

Written by the indexing model from the issue text.