The marginal rejection level is the maximum p-value
in a single tests which still results in rejection of the null
when all other tests have a p-value of 1.
To test the null hypotheses that all p-values are uniform
against a restricted beta family 0 < a <= 1 <= b, the most
powerful pooled p-value linearly combines upper and lower tail
probabilities of the chi-squared distribution with two degrees
of freedom with weights w and (1 - w) where w = (1 - a)/(b - a).
This function estimates the marginal rejection level empirically
by simulating a specified number of null cases to give an empirical
pooled p-value for the rejection level alpha.