`.replace()` behavior on `void` columns

`.replace()` [should operate](https://datatable.readthedocs.io/en/latest/api/frame/replace.html) on columns of types appropriate for the `replace_what` value. At the same time, we [declare](https://datatable.readthedocs.io/en/latest/api/type/void.html) that `void` columns can be used in place where any other type could be expected. However, it is not the case for `.replace()` and replacing missing value with anything doesn't work for `void` columns:

```python
>>> import datatable as dt
>>> DT = dt.Frame([None])
>>> DT.replace(None, 1)
>>> DT
| C0
| void
-- + ----
0 | NA
[1 row x 1 column]
```