Allow some variation in differential entropy analytical tests

[kahaaga]

Test for differential entropy estimators occasionally fail on CI. Because we're using random data from certain probability distributions to estimate the true entropy, the estimates may not be entirely accurate. Currently, it is just tested whether the estimated entropy is equal to the analytical entropy (rounded to two decimal points). However, we should allow some more slack (e.g. 5% or something), so tests don't randomly fail.

This happened for example in #213 .