Have `unpack` support `String` names for the columns

[ablaom]

As in

data = load_iris()
y, X = unpack(==("target")) # now only `==(:target)` is supported now

This can lower the congnitive burden for newcomers familiar with DataFrames.jl which supports string indexing of columns.

[ablaom]

Actually, the implicit design suggested here won't work, as there is no way to know whether the supplied predicates for selection are intended to be applied to strings or symbols. It seems, the only non-breaking solution is to add a keyword argument string_names=false which can be set to true to allow for strings, as in my PR cross-referenced below.

[ablaom]

Another possibility, is to allow, in addition to predicates, symbols, strings, abstract vectors of these, and inverted indices (formed by Not(....) from InvertedIndices.jl). As these things all have testable types, we can just treat any other type as a callable (a predicate) and keep the existing behaviour. So, we can do things like:

v, X, W, y, _ = unpack(table, "column1", [:column2, :column3], Not("column4"), ==(:target))

One cannot in this scheme specify ==("target") but you could do "target".

This would be a non-breaking enhancement that would simplify the common use-cases.

@OkonSamuel What do you think?

Implementation would not be too hard - you simply add at the method front-end a step to convert any of the new supported proxies for a predicate with an actual predicate.

[ablaom]

DataFrames also supports Between(first, last), owned by DataAPI, so perhaps that would need to be supported also.