DEPR: deprecate inplace keyword in the DataFrame.drop() method

[tamhahn]

Greetings!

This PR addresses the tracker at Issue #63207 for DataFrame.drop()

It adds a deprecation warning (Pandas4Warning) for DataFrame.drop(), a deprecation note in docstring, a test for the warning, edits to other affected test files, and whatsnew update.

Adding this warning to DataFrame.drop affected quite a few tests (about 1000 initially failed), so I did 3 things:

1. Added filterwarning decorator to filter out the warning for tests to several test files.
2. Edited _drop_labels_or_levels in generic.py to not use inplace in 2 instances. Other methods (melt and merge) call this and it affected their tests.
3. There was one particular test (test_inplace_drop_and_operation in test_drop.py) where adding the pytest.mark.filterwarning decorator did not filter out the warning, and I made changes to the test itself, which I was trying to avoid. Please let me know if that's not the correct way to address this.

I ran pytest pandas -n 4 -m "not slow and not network and not db and not single_cpu" -r sxX.
The result are the same 7 failures I also get when running that same command on my local main branch (no diffs to upstream).

[tamhahn]

Sorry, a test failed right away. I will look into it and update. Could you please disregard this PR for now.

[jorisvandenbossche]

Looks good!

[jorisvandenbossche]

I think for this test, you could also update the test to not use inplace (from a quick look, I think the test is not really about the inplace drop, but it is using that to construct the expected result of the indexing operation)

[jorisvandenbossche]

2. Edited _drop_labels_or_levels in generic.py to not use inplace in 2 instances. Other methods (melt and merge) call this and it affected their tests.

Yes, good call (we should not be using inplace internally, so those kind of cases can indeed be updated)

Sorry, a test failed right away. I will look into it and update. Could you please disregard this PR for now.

No worries, currently CI is failing for pyarrow and numpy nightly, and that is not related to the changes here and also happening on other PRs

[tamhahn]

Thank you very much, Joris!

I have also removed the use of drop with inplace in the test file you highlighted ([pandas/tests/indexing/multiindex/test_indexing_slow.py as you suggested.

The test did have an assert statement ensuring that df.drop with inplace=true returns None but that was already covered in other tests, so unless there is a special interaction with the MultiIndex-related Issues listed in the test ( # GH7724, GH2646), it is a bit of a redundant test. If I missed it, I'll put it back in.