BUG: Fixed non-unique indexing memory allocation issue with .ix/.loc (GH4280)

[jreback]

closes #4280

Had a weird memory allocation scheme (that's why is a scheme!) when determining
a non-unique indexer. Fixed to use a dynamic schema

In [1]: columns = list('ABCDEFG')

def gen_test(l,l2):
            return pd.concat([ DataFrame(randn(l,len(columns)),index=range(l),columns=columns),
                               DataFrame(np.ones((l2,len(columns))),index=[0]*l2,columns=columns) ])

In [3]: df = gen_test(900000,100000)

In [5]: mask = np.arange(100000)

In [6]: df
Out[6]: 
<class 'pandas.core.frame.DataFrame'>
Int64Index: 1000000 entries, 0 to 0
Data columns (total 7 columns):
A    1000000  non-null values
B    1000000  non-null values
C    1000000  non-null values
D    1000000  non-null values
E    1000000  non-null values
F    1000000  non-null values
G    1000000  non-null values
dtypes: float64(7)

In [7]: df.loc[mask]
Out[7]: 
<class 'pandas.core.frame.DataFrame'>
Int64Index: 200000 entries, 0 to 99999
Data columns (total 7 columns):
A    200000  non-null values
B    200000  non-null values
C    200000  non-null values
D    200000  non-null values
E    200000  non-null values
F    200000  non-null values
G    200000  non-null values
dtypes: float64(7)