[BREAKING] Make DataFrameColumns stop being an AbstractVector

[bkamins]

Fixes #2229
I have added functionalities for DataFrameColumns that I find useful, but please comment if you would like to see more.

[bkamins]

@nalimilan - it should be good to have a look at.

[bkamins]

As usual some corner cases of the changes popped up.
Good change:
DataFrame(eachcol(df)) is now no-op (except copying of columns), previously it dropped column names.
Bad change (but maybe not that bad): eachcol(df)[1,1] now errors as it would produce duplicate column name.

[quinnj]

LGTM; just the use of [begin] causing 1.0 test failures

[bkamins]

Thank you for having a look at it!

[rasmushenningsson]

This looks great!

[bkamins]

@nalimilan - your questions were exactly the things I had to check against Base when implementing it. These interfaces are really non-obvious.

[bkamins]

So the major point of discussion is this case:

julia> x = Any[[1,2,3], [1,2,3]]
2-element Array{Any,1}:
 [1, 2, 3]
 [1, 2, 3]

julia> collect(x)
2-element Array{Any,1}:
 [1, 2, 3]
 [1, 2, 3]

As you can see - I have implemented methods to follow what happens in Base with vectors. We can change it to follow your proposal though. What do you think?

[nalimilan]

OK, makes sense. Then the only remaining issue to settle is whether to create a SubDataFrame parent with multi-indexing.

[bkamins]

You convinced me to return a DataFrame :). I am working on it now and on more flexible firstindex per your comment

[bkamins]

Thank you!