Converting scalar SkyCoord to length-1 SkyCoord array

I have an use case (in photutils) of putting SkyCoord objects in a QTable. However, there is an edge case where the SkyCoord object could either be scalar (only 1 detected source) or contain an array of coordinates (many detected sources). The issue is that scalar objects (including SkyCoord) cannot be added as a table column. So I tried np.atleast_1d(skycoord), but that returns a length-1 ndarray whereas I want a length-1 SkyCoord array (so the table column is always a SkyCoord object regardless of whether it contains 1 or many sources). So my workaround is a bit ugly: SkyCoord(np.atleast_1d(skycoord)):

>>> import numpy as np
>>> from astropy.coordinates import SkyCoord

>>> sc = SkyCoord(10, 10, unit='deg')  # but could also contain an array of coords
>>> sc1 = np.atleast_1d(sc)
>>> sc1  # not what I want
array([<SkyCoord (ICRS): (ra, dec) in deg
    (10., 10.)>], dtype=object)

>>> sc2 = SkyCoord(np.atleast_1d(sc))
>>> sc2  # good!
<SkyCoord (ICRS): (ra, dec) in deg
    [(10., 10.)]>

sc1 and sc2 produce different table columns:

>>> from astropy.table import QTable
>>> tbl = QTable()
>>> tbl['col1'] = sc1
>>> tbl['col2'] = sc2
>>> tbl
                       col1                             col2  
                                                          deg,deg 
-------------------------------------------------- ---------
<SkyCoord (ICRS): (ra, dec) in deg
    (10., 10.)>                                     10.0,10.0

I always want the column to be a SkyCoord object (in the case of 1 or many objects).

I discussed this a bit offline with @eteq and he suggested adding a atleast_nd(dim) method to SkyCoord objects to turn a scalar SkyCoord into a one-element n-dimensional SkyCoord array (I like this suggestion).

[bsipocz]

Oh, sure. But that part is unlikely to get resolved this week, so I removed the milestone.

[mhvk]

Arrggh, that reminds me of #8610 - really would be lovely if np.atleast_1d just worked, and with __array_function__, it can…