Support computation of alpha for subsamples

[johannabi]

I would like to compute alpha values for different subsets of my annotated data. But computing alpha for each subset separately needs different data for the random_coincidences matrix. Since the random_coincidences matrix is computed based on the reliability_data, the expected values get skewed if annotations in subsets are unbalanced.

It would be nice if one could either add all_reliability_data as parameter or if one could compute random_coincidences by oneself and add this matrix as parameter to the alpha-function.