Fix flaky relative binning tests
Sometimes the relative binning accuracy checks fail (e.g., here), it looks like no random seed is set for those tests.
We should find a random seed where this passes, I think it is expected that this approximation should sometimes not surpass the threshold.