Add vignette explaining the use of `prior=` argument to correct for biases in data

[pglpm]

For variates that have been controlled during sampling – that is, whose frequencies are more or less chosen by the researcher – one cannot directly calculate conditional probabilities using Pr(Y = [controlled variate], X = ...), because the results would be biased. But Inferno allows one to calculate these probabilities by correcting the bias of the controlled frequencies, through the argument prior= in Pr().

Add vignette explaining this situation and the use of this argument. An example could be with the 'island' variate in the Penguins example.