BUG: Non-unique indexing via loc and friends fixed when slicing (GH3659_)

[jreback]

closes #3659

This is if you try a non_monotonic selection on a non_unique index (a mouthful)!
The reason is we cannot determinate a proper start/end point on what to include

In [11]: df = DataFrame({'A' : [1,2,3,4,5,6], 'B' : [3,4,5,6,7,8]}, 
           index = [0,1,0,1,2,3])

In [18]: df
Out[18]: 
   A  B
0  1  3
1  2  4
0  3  5
1  4  6
2  5  7
3  6  8

In [12]: df.loc[1:]
KeyError: 'cannot perform a slice operation on a non-unique non-monotonic index'

On a non_unique, but monotonic index, however, slicing works normally
(notice, since we are using loc, that both endpoints ARE included)

In [13]: df = DataFrame({'A' : [1,2,3,4,5,6], 'B' : [3,4,5,6,7,8]}, 
       index = [0,1,0,1,2,3]).sort(axis=0)

In [14]: df
Out[14]: 
   A  B
0  1  3
0  3  5
1  2  4
1  4  6
2  5  7
3  6  8

In [15]: df.loc[1:]
Out[15]: 
   A  B
1  2  4
1  4  6
2  5  7
3  6  8

In [16]: df.loc[1:2]
Out[16]: 
   A  B
1  2  4
1  4  6
2  5  7

[jreback]

@wesm pls give this a once over

[jreback]

I was making this more complicated that it was. Solution is sort of trivial (just needed to raise an exception when get_loc returns a boolean array rather than a slice, indicating index is non-monotonic and hence not slice indexable)

that is a mouthful