deal with pandas new `nan_is_na` mode

[MarcoGorelli]

In pandas 3.0 there will be a nan_is_na mode: pandas-dev/pandas#62040

If it's True (default), then nan will be indistuinguishable from na. If False, they will be distinguished, and isna will only report in NA values

We need to make a decision here about what to do for pandas-like

I think what would be most natural would be to just keep doing what we're doing:

• for nan, return whatever s != s outputs
• for na, return whatever s.isna() outputs

Note that a consequence of this is that when nan_is_na is True, then s_nw.is_nan() would differ from s_nw.is_null():

In [1]: import pandas as pd

In [2]: pd.set_option("mode.nan_is_na", True)

In [3]: s = pd.Series([-1, 1, None], dtype='Int64') ** .5

In [4]: s
Out[4]: 
0    <NA>
1     1.0
2    <NA>
dtype: Float64

In [5]: s.isna()
Out[5]: 
0     True
1    False
2     True
dtype: bool

In [6]: s != s
Out[6]: 
0     <NA>
1    False
2     <NA>
dtype: boolean

but I think that's OK, if pandas wants to mask NaN values to be NA ones, then that's their decision

[dangotbanned]

It seems there's an active discussion again (as of yesterday) on the PR that introduced this change:

• API: mode.nan_is_na to consistently distinguish NaN-vs-NA pandas-dev/pandas#62040 (comment)