BUG: replace of numeric by string / dtype coversion

[jreback]

xref #12747

This was an item not closed by #12780 (but listed in #12747)

In [2]: pd.Series([1,2,3]).replace('1', np.nan)
Out[2]: 
0    NaN
1    2.0
2    3.0
dtype: float64

[ucals]

Output should be:

0    1.0
1    NaN
2    3.0
dtype: float64

right? I'd like to try this one

[jreback]

The result should be the same as this. IOW the string '1' is NOT found in the original dataset (its a numeric 1). So should return unchanged. Step thru the code and you will see the modification happening, this is numpy behavior, but its not normal / natural in the real world :>

In [2]: pd.Series([1,2,3]).replace('10', np.nan)
Out[2]: 
0    1
1    2
2    3
dtype: int64