Grouping over PeriodIndex breaks its format

[leonbaum]

With the master branch (f014b01), grouping over a PeriodIndex is breaking the format of the index:

In [100]: df = pandas.DataFrame({'x': [1, 2, 3], 'y': ['2012-12-18', '2012-12-19', '2013-01-04']})

In [101]: df['y'] = df.y.astype('datetime64[D]')

In [102]: df = df.set_index('y').to_period('M')

In [103]: df
Out[103]: 
         x
2012-12  1
2012-12  2
2013-01  3

In [104]: df.groupby(level=0).sum()
Out[104]: 
     x
515  3
516  3

I would expect to still see the dates in the index.

[changhiskhan]

Now Index(PeriodIndex) returns PeriodIndex if dtype is unspecified