Datatypes of booleans with missing values in pandas

Question

An example:

import pandas as pd
import numpy as np

df  = pd.DataFrame({'A':[True, False, True],
                    'B':[True, False, np.nan],
                    'C':[True, False, True]})
df.loc[[2],'C'] = np.nan

print(df, df.dtypes, sep='\n\n')

>>>

       A      B    C
0   True   True  1.0
1  False  False  0.0
2   True    NaN  NaN

A       bool
B     object
C    float64
dtype: object

I get that in "C", the datatypes get converted to float (and not int b/c ValueError: cannot convert float NaN to integer). But why doesn't the same happen in "B"? Why can the bool type column be converted into float64 when the data are edited, but a column of Booleans with some missing data can't be converted?

I've run into some operations in pandas that I expect to work on column "A", but I end up needing to be more explicit about the datatype (here for example).