BUG: multilevel indexing with new values

[jreback]

In [2]: df = pd.DataFrame(data=[[0, 2, 1], [np.nan, 5, 4]], index=pd.date_range('2014-01-01', '2014-01-2', name='timestamp'), columns=pd.MultiIndex.from_product([['value'], [24, 3766, 5061]], names=[None, 'sid']))

In [3]: df
Out[3]: 
           value          
sid         24   3766 5061
timestamp                 
2014-01-01     0    2    1
2014-01-02   NaN    5    4

In [4]: assets = [3766,5061,123]

In [5]: df.reindex(columns=assets,level=1)
Out[5]: 
           value     
sid         3766 5061
timestamp            
2014-01-01     2    1
2014-01-02     5    4

So this should include 123 as the last column (and all nans) as well.
The indexer returned in np.array([1,2]), but should be np.array([1,2,-1]), see here

only 1 test I could find actually tests this behavior, see here

but only tests that the SAME values are returned, not less or more (which is the point of this issue)

[jreback]

superseded by #12319