BUG: Fixes issue #3334: brittle margin in pivot_table.

[guyrt]

closes #3334 Brittle pivot margins

Issue was that pivot tables that use all columns of the original DataFrame in rows and cols failed on the margin computation. This is a special case for margins: one should use the index itself as the value.

>> df = DataFrame({'Response' : ['Y', 'N' ,'N', 'Y', 'Y', 'N'],
            'Type' : ['A', 'A', 'B', 'B', 'B', 'C']})

>> pivot_table(df, rows='Response',cols='Type',aggfunc=len,margins=True)
>> Type      A  B   C  All
>> Response               
>> N         1  1   1    3
>> Y         1  2 NaN    3
>> All       2  3   1    6