FTRL fails when the target column stype changes after fit

```
>>> from datatable.models import Ftrl
>>> from datatable import dt
>>> ft = Ftrl()
>>> DT = dt.Frame(["1", "2", "3"])
>>> res = ft.fit(DT, DT)
>>> DT = dt.Frame([1])
>>> res = ft.fit(DT, DT)
TypeError: Column label of type int8 in the left Frame cannot be joined to column label of incompatible type str32 in the right Frame
```

We should either allow this, or explicitly forbid by issuing more clear error message.