BUG: np.min on DataFrame with a non-ordered Categorical col

Discovered while debugging the try/excepts in groupby:

```
cat = pd.Categorical(['a', 'b', 'c', 'b'], ordered=False)
ser = pd.Series(cat)
df = ser.to_frame()
>>> np.min(cat)   # <-- correctly raises
>>> np.min(ser)   # <-- correctly raises
>>> np.min(df)     # <-- incorrectly returns Series(['a'])
```
