BUG: clip with a list-like threshold with a nan is broken

[jreback]

xref #19976

.clip is basically a single or double signed where, where there is a nan present we just drop the bound entirely which is not correct. A fix for this would also fix #19976. So [31] should yield [30], however for scalars we ignore the nans, so maybe could do that here.

In [26]: s = pd.Series([1,2,3])

In [27]: s
Out[27]: 
0    1
1    2
2    3
dtype: int64

In [28]: s.where(s>4,4)
Out[28]: 
0    4
1    4
2    4
dtype: int64

In [29]: s.clip(4)
Out[29]: 
0    4
1    4
2    4
dtype: int64

# broken
In [30]: s.where(s>[0,4,np.nan],[0,4,np.nan])
Out[30]: 
0    1.0
1    4.0
2    NaN
dtype: float64

In [31]: s.clip([0,4,np.nan])
Out[31]: 
0    1
1    2
2    3
dtype: int64

# ok
In [2]: s.where(s>[0,4,4],[0,4,4])
Out[2]: 
0    1
1    4
2    4
dtype: int64

In [3]: s.clip([0,4,4])
Out[3]: 
0    1
1    4
2    4
dtype: int64