BUG: dask.dataframe fill_value doesn't seem to do anything...

Try the following code... it should result in a dask dataframe with no `NaN` values, but alas, it still shows up. I guess `fill_value` isn't working. 

```
import dask.dataframe as ddf
import pandas as pd
import numpy as np

x = np.arange(1, 10, dtype=float)
x[5] = np.nan

x = pd.DataFrame(x, columns=['foo'])
x = ddf.from_pandas(x, chunksize=3)
x.pow(2, fill_value=1).compute()
```
