Enable the predicate pushdown pytest affected by Arrow v19 stats incompatibility bug.

[mhaseeb123]

Enable the disabled portion of the test_parquet_bloom_filter pytest which aborts due to an Arrow v19 incompatibility issue apache/arrow#45283 with stats.

The arrow bug has been fixed by PR apache/arrow#45285 but there isn't yet an Arrow release containing this PR. Once there is one available and cuDF is bumped to it, we can revert the said pytest back to the following:

def test_parquet_bloom_filters(
    datadir, stats_fname, bloom_filter_fname, predicate, expected_len
):
    fname_bf = datadir / bloom_filter_fname
    df_bf = cudf.read_parquet(fname_bf, filters=predicate).reset_index(
        drop=True
    )

    fname_stats = datadir / stats_fname
    df_stats = cudf.read_parquet(fname_stats, filters=predicate).reset_index(
        drop=True
    )

    # Check if tables equal
    assert_eq(
        df_stats,
        df_bf,
        )
    # Check for table length
    assert_eq(
        len(df_bf),
        expected_len,
    )

See comment by @mhaseeb123 in #17422 (comment) for more details.

[vyasr]

@mhaseeb123 I think this should be unblocked now if we bump to testing against the latest version of arrow.

[mhaseeb123]

@mhaseeb123 I think this should be unblocked now if we bump to testing against the latest version of arrow.

Thanks, I can take a look at this soon.

[mhaseeb123]

Looks like @bdice beat me to updating Arrow bounds in #19592. I will enable the relevant pytests once it merges

[vyasr]

#19592 was merged, so this work is unblocked now.

[mhaseeb123]

@vyasr Looking at the skip condition

cudf/python/cudf/cudf/tests/input_output/test_parquet.py

Lines 4443 to 4446 in 80d8c2c

	@pytest.mark.skipif(
	pa.__version__ == "19.0.0",
	reason="https://github.com/rapidsai/cudf/issues/17806",
	)

it looks like the test should run now as is since it's only disabled when pa version == "19.0.0. We can still keep this skip condition to avoid any inadvertent errors.

I am thinking of updating the reason string to point to the Arrow issue apache/arrow#45283 due to which we disabled this test and close this issue. What do you think?

[vyasr]

Yes, I think updating the reason to include both links would be good, then we can close. Thanks!