Stop beads from being erroneously removed after inference

[gesinecauer]

torm is a filter that removes beads with 0 reads. The lower half of counts, as well as the diagonal, are 0. Therefore counts needs to be added to counts.transpose() for this sum to work as intended.

[gesinecauer]

I see that you already resolved a similar problem that was happening in iced.filter.filter_low_counts using X_sum = (np.array(X.sum(axis=0)).flatten() + np.array(X.sum(axis=1)).flatten()), which is actually probably a better way to do this

[gesinecauer]

Also, the same thing needs to be done for PM2, MDS, and NMDS - forgot to make the changes for those as well