Logging is very slow

During #128 (closed) and #164 (closed), the actual decoding takes ~6 seconds. Processing the main log callback adds very little overhead. Getting to the point of being ready to dump into the Python logging system adds ~2s. Actually doing it adds another ~4s. So that is fully an additional 100% for logging the same message thousands of times.

Perhaps we should severely raise the default logging threshold to ERROR.

It would be interesting to see if we could instrument the loggers so that the C callback is aware of the minimum Python threshold.