API: DatetimeIndex.asobject raises ValueError when contains NaT

[sinhrks]

Related to #7485.

Currently, DatetimeIndex.asobject raises ValueError when it contains NaT, otherwise PeriodIndex doesn't inconsistently.

I think these should work the same way, and would like to confirm PeriodIndex should raise ValueError, or DatetimeIndex.asobject allows NaT. There may be historical reason, but I'm not sure why DatetimeIndex raises ValueError even though we can create object-dtype Index containning NaT

didx = pd.DatetimeIndex([datetime.datetime(2014, 1, 1), pd.NaT])
didx.asobject
# ValueError: DatetimeIndex with NaT cannot be converted to object

pidx = pd.PeriodIndex(['2014-01', 'NaT'], freq='M')
pidx.asobject
# Index([2014-01, NaT], dtype='object')

[jreback]

I think here's the reason. Long time ago datetimes were object dtype when embeded in a frame (prior 0.10.0). Now only a dateitme64[ns] with a tz and PeriodIndex need object dtype.

So I think ok to fix DatetimeIndex to allow an object dtype output (though issue is that it may try to coerce it back to datetime64[ns]! in the constructor)

See here for the embedding code: https://github.com/pydata/pandas/blob/master/pandas/core/frame.py#L2047 (should prob change PeriodIndex to use the same type of _to_embed method as well).