BUG: groupby.std with no numeric columns and numeric_only=True raises

```
df = pd.DataFrame({'a': list('xyz'), 'b': list('def')})
gb = df.groupby('a')
print(gb.std(numeric_only=True))
```

raises `TypeError: All columns were dropped in grouped_reduce`. Instead, it should return an empty frame with index consisting of the groups.