arrow

ARROW-13158: [Python] Fix StructScalar contains and repr with duplicate field names

This PR does several small things:

- Fix `__contains__` to not check the actual keys (and not rely on `__getitem__` since that also supports integers)
- Add tests for the case of StructScalar with duplicate field names, and ensure some behaviours that already worked (`__getitem__` with integers, `__iter__`, `keys()`)
- Changed `list(scalar)` (`__iter__`) to also return the keys for a "null" scalar (since we do allow `s["key"]` in that case, that seems most consistent?)
- Changed the `repr` to use a tuple instead of dict representation. I know this just postpones a bit the `as_py()` discussion (should it return dict vs tuple?), but at least this ensures that the `repr` doesn't fail on the short term, and so you can inspect the object.
- I added an explicit `items()` method (overriding the ones from Mapping), because the return value of this can in theory support duplicate fields, and this can be a way to get a list of tuples already. But, this doesn't fully follows the Mapping API as the return value is different (not a `dict_items` object).

Closes #10591 from jorisvandenbossche/ARROW-13158

Authored-by: Joris Van den Bossche <jorisvandenbossche@gmail.com>
Signed-off-by: Antoine Pitrou <antoine@python.org>