[ntuple] Add RPageStorage::Close

[silverweed]

This Pull request:

adds a virtual Close method to RPageStorage, to be used by pythonizations of RNTupleReader and RNTupleWriter.
Also updates those pythonizations to use it.

Checklist:

• tested changes locally
• updated the docs (if necessary)

[vepadulano]

Thanks, this is great! Two remaining comments from my side:

• It would be nice to add also a simple test for the reader and writer
• What happens if I call reader.Close() and then somehow try to access it again? From the changes it's not immediately clear to understand if I get an error/info or there could be unexpected behaviour.

[github-actions]

Test Results

    22 files      22 suites   3d 11h 43m 37s ⏱️
 3 770 tests  3 769 ✅ 0 💤 1 ❌
74 989 runs  74 988 ✅ 0 💤 1 ❌

For more details on these failures, see this check.

Results for commit 115bcb0.

[pcanal]

LGTM.

[silverweed]

Rethinking about this, I now think this is not the best solution.

As far as I see it, the Close method is not really needed on the C++ side and it's just a side effect of Python needing a "valid but invalid" object after the context manager exits. If we go and add this to the C++ side we will end up with additional unwarranted code complexity stemming from the fact that the Reader and Writer lose the invariant that the source/sink is valid as long as they are valid, which will result in adding validity checks around which would not be needed in principle.

I think the proper solution would be to do this purely on the Python side: in the __exit__ method we reset the internal pointer to the reader/writer (assuming this is possible) and hijack the proxy object to raise an exception whenever you try to access a method of the underlying object.

@vepadulano @jblomer opinions on this?