Implement dimension-aware skipmissing

[stillyslalom]

I added a method skipmissing(v, dims) = mapslices(skipmissing, v, dims) which allows compact notation for skipmissing over multidimensional arrays. Can be composed with basic stats functions to provide functionality similar to Matlab & Numpy's nanmean(array, dim), etc:

julia> mean.(skipmissing([1 missing; 3 4], 1))
1×2 Array{Float64,2}:
 2.0  4.0

This generates lots of views & iterators, so it's not the most performant, but it's a decent start. ref

[nalimilan]

Thanks. Actually, I think we'll need a more complex implementation where you will call sum(x; dims=...) directly on the SkipMissing object. That would be more consistent with the pattern according to which dims is passed to the reduction function, it would be more natural than returning an array of SkipMissing objects, and more importantly maybe it would be more efficient.

I've already a branch where all of this mostly works, but I need to sort out some details.