Allow for scalars in a grouped @transform

[pdeffebach]

Resolves issue #98 with a one-liner.

This way we can spread grouped vector-to-scalar operations across groups easily. This behavior is already feasible, but it threw an error with strings because eltype(String) is Char. Now it will allocate a vector of type String.

Next step is to automatically expand the type of the vector if a group is entirely missing. Note that this would have thrown an error before this PR as well.

[bkamins]

Actually given https://discourse.julialang.org/t/best-way-to-append-vectors-of-various-types/12837 maybe we should fix also this in one shot?

This would require precomputing all results and finding their common type. DataFrames.jl uses internal promote_col_type to achieve it.

[pdeffebach]

Closed in favor of #101 This has that implementation.