Interchange protocol: standardised way of getting back the native object

[MarcoGorelli]

If someone wants a standardised way of selecting columns 'a' and 'b' using the interchange protocol, then it would be nice to be able to do something like

df.__dataframe__().select_columns_by_name(['a', 'b']).to_native()

pandas and Polars both use ._df for this:

In [24]: pl.DataFrame({'a': [1,2,3], 'b': [4,5,6], 'c': [7,8,9]}).__dataframe__().select_columns_by_name(['a', 'b'])._df
    ...:
Out[24]:
shape: (3, 2)
┌─────┬─────┐
│ a   ┆ b   │
│ --- ┆ --- │
│ i64 ┆ i64 │
╞═════╪═════╡
│ 1   ┆ 4   │
│ 2   ┆ 5   │
│ 3   ┆ 6   │
└─────┴─────┘

In [25]: pd.DataFrame({'a': [1,2,3], 'b': [4,5,6], 'c': [7,8,9]}).__dataframe__().select_columns_by_name(['a', 'b'])._df
    ...:
Out[25]:
   a  b
0  1  4
1  2  5
2  3  6

but that's not standardised by the interchange protocol

Use-case: the interchange level of support which we have in Narwhals: narwhals-dev/narwhals#1263

[MarcoGorelli]

closing as dupe of #85