BUG: replace int64 dtype with bool coerces to int64

[sinhrks]

# NG
s = pd.Series([1, 2, 3])
s.replace(2, True)
#0    1
#1    1
#2    3
# dtype: int64

If target has object dtype, it works.

s = pd.Series(['x', 2, 3])
s.replace(2, True)
#0       x
#1    True
#2       3
# dtype: object

[jreback]

This is numpy casting at work. If you can fix in a performant way, then ok.

[mroeschke]

Looks fixed on master. Could use a regression test.

In [26]: pd.__version__
Out[26]: '0.26.0.dev0+490.g9cfb8b55b'

In [27]: s = pd.Series([1, 2, 3])
    ...: s.replace(2, True)
Out[27]:
0       1
1    True
2       3
dtype: object