API: replace casting dtype instead of converting NA value when replacing missing value with another missing value

[phofl]

See discussion in #48234

when doing

pd.Series([1.0, np.nan]).replace(np.nan, value=None)

this casts to object, which is inconsistent with what other methods do.

The non special logic would look like:

def replace(self, old, new):
    if is_valid_na_for_dtype(new, self.dtype):
        new = self. _standardize_fill_value(new)

cc @jbrockmendel @mroeschke

[jbrockmendel]

I think if the user explicitly asks for .replace(nan, None) we should give it to them. Isn't the intent unambiguous?