fix select bug with copycols=false on SubDataFrame

[bkamins]

Fixes the following bug:

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

julia> dfv = view(df, :, :)
3×1 SubDataFrame
 Row │ a
     │ Int64
─────┼───────
   1 │     1
   2 │     2
   3 │     3

julia> select(dfv, x -> true, copycols=false) # bug - this should error
3×1 SubDataFrame
 Row │ a
     │ Int64
─────┼───────
   1 │     1
   2 │     2
   3 │     3

julia> select(dfv, x -> true) # correct behavior in case of copying, but impossible to achieve without copying when working on a view
3×1 DataFrame
 Row │ x1
     │ Bool
─────┼──────
   1 │ true
   2 │ true
   3 │ true

[bkamins]

Also fixed:

julia> select(df, 1.0)
ERROR: ArgumentError: Unrecognized column selector 1.0 in AsTable constructor
... (here removed a long stack-trace) ...
caused by: MethodError: no method matching getindex(::DataFrames.Index, ::Float64)

by

julia> select(df, 1.0)
ERROR: ArgumentError: invalid index: 1.0 of type Float64

which I think is nicer

[nalimilan]

This will break code that incorrectly relies on getting no error, so maybe it could wait for 1.5 to limit disruption?

[bkamins]

This will break code that incorrectly relies on getting no error

I do not see what correct code would be broken here.

The only possibility that code worked that currently fails is select(dfv, fun, copycols=false), where fun is just a function (taking a data frame as an input). In such a case no transformation is done but the function currently is wrapped in Cols.
This happens in this line view(dfv, :, Cols(newinds...)), which makes sense ONLY if newinds is actually a vector of column selectors only.
However, it sometimes works in the old code, as in the example:

julia> select(dfv, x -> true, copycols=false) # bug - this should error
3×1 SubDataFrame
 Row │ a
     │ Int64
─────┼───────
   1 │     1
   2 │     2
   3 │     3

because Cols accepts taking a function as its argument (which probably was not the case when this code was written 😞, as we added this feature later to Cols is I remember correctly). But it changes its meaning. Now the function is not passed SubDataFrame, but is a predicate indicating which column should be kept (in the above example x -> true is interpreted as keep all columns selector).

For me it is a bug that never can produce what user wanted. And that is why I thought to make a patch release for it.

[nalimilan]

OK, fair enough!

[bkamins]

Thank you!