Better support for skipmissing

[pdeffebach]

I'm surprised things aren't set to work with any iterator. Am I doing something wrong?

x1 = [1, 2, 3, 4, missing]
x2 = [1, 1, 2, 3, missing]
julia> ApproximateTwoSampleKSTest(skipmissing(x1), skipmissing(x2))
ERROR: MethodError: no method matching ApproximateTwoSampleKSTest(::Base.SkipMissing{Array{Union{Missing, Int64},1}}, ::Base.SkipMissing{Array{Union{Missing, Int64},1}})
Closest candidates are:
  ApproximateTwoSampleKSTest(::Any, ::Any, ::Any, ::Any, ::Any) at C:\Users\Pdeffebach\.julia\dev\HypothesisTests\src\kolmogorov_smirnov.jl:143
Stacktrace:
 [1] top-level scope at none:0

Would be great to widen the type signatures where needed.

[nalimilan]

Just like in StatsBase, this is probably restricted mostly for historical reasons, but in some cases accepting any iterator may require refactoring code a bit.

In the case of tests taking multiple arrays, there's also the problem that missing values are unlikely to appear in the same place in both inputs. So it may be better to use df = DataFrame(x1=x1, x2=x2); sdf = view(df, completecases(df)); Test(df2.x1, df2.x2). See also JuliaLang/julia#30596.

[pdeffebach]

Yeah I should get around to that PR. KS tests don't have any order attached to them, but presumably some tests here that do? Time series tests raise a number of issues I guess.