Add broadcasting of selectors to the minilanguage

[bkamins]

@nalimilan + @pdeffebach: I am sharing the preview of the functionality making e.g. Not(:x1) .=> sum work in the minilanguage.

The core mechanism should be working for intended cases.

What is todo:

• implement expansion engine
• inject the inner mechanics to select, subset etc. functions that should call it
• add tests of intended use cases
• add tests for target functions
• add tests of corner cases (i.e. to check if we are good if we pass something that is not intended)
• add NEWS.md
• add docstring
• add documentation

[bkamins]

Additionally I have caught two bugs in the existing implementation:

julia> df = DataFrame(a=1,b=2,c=3)
1×3 DataFrame
 Row │ a      b      c
     │ Int64  Int64  Int64
─────┼─────────────────────
   1 │     1      2      3

julia> select(view(df, 1:1, 1:2), [:x1, :x2] .=> [sum length])
ERROR: ArgumentError: Unrecognized column selector: Pair{Symbol, _A} where _A[:x1 => sum :x1 => length; :x2 => sum :x2 => length]

julia> select(groupby(df, :a), Symbol[] .=> [sum length])
ERROR: MethodError: no method matching select(::GroupedDataFrame{DataFrame}, ::Matrix{Any})

(both are fixed)

[nalimilan]

I guess this is again because of broadcasting returning Matrix{Any}? It's somewhat ugly to have to handle this special case here. Would there be any way to fix this elsewhere?

[bkamins]

Yes - this is the Matrix{Any} case special handling. This is essentially the only place we can handle this in a short way. The point is that we want to do AbstractVecOrMat{<:Pair} check below (not to allow arbitrary matrices to leak further unless they are empty as then it does not matter what eltype they have - and thus the additional check).

[bkamins]

CI failure here is unrelated.

[nalimilan]

There's a syntax problem here. Just copy the text from above?

[bkamins]

I have forgotten to sync the docstrings. Fixed.

[bkamins]

Thank you!