Make poscounts accessible from .deseq2()

[BorisMuzellec]

Is your feature request related to a problem? Please describe.

DeseqDataSet.fit_size_factors() has a fit_type=poscounts option to take the average on positives reads only, but it is neither accessible when initializing a DeseqDataSet or when running .deseq2().

This means it can only be used by calling dds.fit_size_factors(fit_type=poscounts) manually.

Describe the solution you'd like

Add a size_factor_fit_type argument to either DeseqDataSet (in which case it would also become an attribute) or .deseq2()

Additional context
I'm wondering whether poscounts would be a better fall-back option than iterative, given how buggy the iterative mode is (cf #198).

This would introduce another difference compared with DESeq2 though.