Add worst-case samplesize calculation for normal tests

[auduno]

Similarly to bernoulli comparison of proportions we would like to know the worst-case samplesize at a given level. This can not be simulated, but needs to be calculated analytically.