Performance for `df.items` and other iterators

Looping over the Modin dataframe is slow when you use `df.items` and other iterators. Part of this is because we are still materializing the data in the driver. This does not scale.

Since we have better indexing performance, we should take advantage of that here.