Remove or unittest index cast at the end of the feature extraction

[nils-braun]

Are we testing the types of indices that need this recasting in some unit tests?
If not, do we actually need this line of casting?

Originally posted by @MaxBenChrist in #705

[dbarbier]

IMHO this is the exact same problem as with sort_index; current code looks good, it seems natural to return a DataFrame with an index identical to df[column_id]. I do not see why this should be changed or tested, there is no test to ensure that index is sorted.