A bug in ealy stopper

[Minorant]

Thanks for your excellent work.
I noticed that in llmpebase/examples/BoTReasoning/early_stopper.py, the stop_via_comment function is used to determine whether a comment is a positive evaluation.
However, it’s clear that conclusions such as “no errors” or “no mistake”, which contain double negatives, will lead to incorrect results. This issue is especially evident in DeepSeek Chat, and it can cause the process to keep looping until all iterations are exhausted.