Enable back detection of errors in logs when running integration tests

We decided a while ago in https://forum.xwiki.org/t/remove-log-validation-for-functional-tests/6026 to stop failing tests in case of error or warning logs because we were not managing to handle the false positive. However, we recently found at least one regression that could have been spotted by that mechanism.
So we should enable it back at least to capture error logs in integration tests.