Sum duplicate rows in muon.atac.tl.count_fragments_features

[benjamin-james]

Is your feature request related to a problem? Please describe.
When I am converting range counts to smaller integers (i8 to i1, for example), sums from count_fragments_features can overflow, even though all numbers are below the maximum value supported. Apparently many duplicate row, column pairs can be inserted in count_fragments_features which can lead to this error.

Describe the solution you'd like

mx = mx.tocsr().transpose()
mx.sum_duplicates()
return AnnData(X=mx, obs=adata.obs, var=features)

Describe alternatives you've considered
Simply calling sum_duplicates() after count_fragments_features is what I currently do.

Additional context
I generally use count_fragments_features to count peak sets from a fragments file, so finding correct values so that they can be clipped per-peak is important.