Should allow multifactor labels to be used in sampling

[oxinabox]

In particular, this required to allow useful use of the targetfun.

Added more tests to that effect

[Evizero]

I am unsure about this. It looks to me as if this change simply hides an implicit use of obsview. I tend to think we should require a user to be specific about it and instead call

# EDIT: nope
oversample(obsview(src); targetfun=x->x[1]>x[2])

The implicit obsview seems scary because then the result looks very different than the input

Edit: misread the code

[Evizero]

Or alternatively

julia> oversample(src; targetfun=x->x[1,:].>x[2,:])
2×2578 SubArray{Int64,2,Array{Int64,2},Tuple{Colon,Array{Int64,1}},false}:
 1  3  4  4  4  2  2  2  2  2  4  3  …  2  3  4  3  1  3  3  4  4  4  2  4
 3  1  3  3  2  2  4  3  3  3  3  4     4  3  2  4  4  2  3  3  2  3  2  2

[Evizero]

Took me a while but now I think I understand what is going on and I think I like it. This could be a really good change.

I have to think about it a little to wrap my head around. For example,

• what would it mean for custom types, such as DataFrame?
• does this take away flexibility when having custom subset types?
• any other caveats
• does this slow down things?

[Evizero]

Lets go with your change. I'll experiment with it locally