BUG: wrong type is returned by iat[] for date time series

These return object of type class `pandas.tslib.Timestamp`:

```
    print type(pd.Series(['2014-01-01', '2014-02-02'], dtype='datetime64[ns]')[0])
    print type(pd.Series(['2014-01-01', '2014-02-02'], dtype='datetime64[ns]').min())
```

but using `iat[]` returns object of type type `numpy.datetime64`:

```
    print type(pd.Series(['2014-01-01', '2014-02-02'], dtype='datetime64[ns]').iat[0])
```
