improved nan treatment on summing

[JGuetschow]

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

When using ds.pr.sum with skipna=True on data which contains NaN the result is currently 0 if all data points summed over are nan. I usually need functionality where the result is NaN when all input are NaN but if there are one or more non-NaN values the result is the sum of the non-NaN values .

Describe the solution you'd like

Use min_count=1 in the call of ds.sum (in the definition of ds.pr.sum) when skipna=True.

Alternatives
Always pass min_count=1 to ds.pr.sum when using it.

Would there be any problems in existing functionality when introducing this change @mikapfl ?

[mikapfl]

I don't think this would impact existing functionality and sounds like a sensible default.