Let DataFrame behave more like GroupedDataFrame with one zero-key group

[jariji]

df::DataFrame is conceptually similar to groupby(df, []), so keys(df) could work but currently throws a MethodError. This change would simplify the mental model of using these objects.

[bkamins]

We cannot do it.

The contract for keys is:

  keys(iterator)

  For an iterator or collection that has keys and values (e.g. arrays and dictionaries), return an iterator over the
  keys.

AbstractDataFrame is not an iterator (as opposed to GroupedDataFrame). Also I do not think we want to make it an iterator with a single (or no) groups.

I initially was open to discuss it, but it seems to me that it is almost impossible that we will change this decision (people would be confused if we made AbstractDataFrame iterable this way), so I am closing the issue. But maybe @nalimilan will have another opinion.