Support Pandas 2

[Flix6x]

This should get us ready for upgrading to Pandas 2.

The good people of sktime are still working on compatibility with Pandas 2. Their latest release (0.17.1) already resolves any problems caught by our tests, but their pyproject.toml still suggests pandas<2.0.0.

[Flix6x]

Good to know: our main test suite doesn't rely on sktime and in this PR automatically tested against pandas==2.0.0. Our forecast test suite does rely on sktime and tested against pandas==1.5.3.

[nhoening]

Any tips on reviewing this?
Should I consult the 2.0 release notes?

[Flix6x]

No tips really.

I was happy to have our tests pass. Let me know if I need to check against breaking changes explicitly.

[nhoening]

Which tests? Only the ones here, or also FlexMeasures?

[nhoening]

Maybe we can use this PR to update the Python version we support, as well? This is the highest:

"Programming Language :: Python :: 3.9",

But I guess 3.10 or even 3.11 are also supported.

[Flix6x]

Which tests? Only the ones here, or also FlexMeasures?

Just the ones here. FlexMeasures still needs some attention, but that's a separate issue (and different repo), right?

[Flix6x]

Maybe we can use this PR to update the Python version we support, as well?

We test 3.10, but not 3.11. So I think I'd just add 3.10 for now.

[nhoening]

Well it would add confidence here. But of course if it requires more work there, we need to see how we can manage.

[Flix6x]

I tried it out (FM with Pandas 2), and found 1 more issue that timely-beliefs tests should have caught. I'll expand the test suite accordingly, and implement a fix (or workaround, because I believe it might be a regression in Pandas).

[nhoening]

You will do that here in this PR right?

[Flix6x]

It turned out that the DataFrame constructor still needs to call finalize in order to propagate the metadata, so I reverted the version check for said constructor.

It is only the Series constructor that can no longer call finalize, otherwise I get a RecursionError: maximum recursion depth exceeded while calling a Python object.

[nhoening]

Looks good, but I have one question.