Ensure that the CDS unit parser can deal with '[-]' - #11250

This introduced a regression into the astroquery tests. Probably not a big deal and we have to fix it in our tests, but others may run into it, too:

Prior to this:

>>> from astroquery.imcce import Miriade
>>> result = Miriade.get_ephemerides('3552', coordtype=5)
>>> result['airmass']
UnrecognizedUnit(-)
>>> result['airmass'].unit == '-'
True
>>> result['airmass'].unit.find_equivalent_units()
AttributeError: 'UnrecognizedUnit' object has no attribute 'find_equivalent'

After:

>>> result['airmass']
Unit("---")
>>> result['airmass'].unit.find_equivalent_units()
  Primary name | Unit definition | Aliases
[
  percent      | 0.01            | pct     ,
]

The raw votable returns can be accessed as e.g. (which is then parsed with astropy.io.votable.parse):

from io import BytesIO
response = Miriade.get_ephemerides_async('3552', coordtype=5)
voraw = BytesIO(response.content)

Is Unit("---") supposed to resolve to percent? 👀

Unit('---') is dimensionless, so percent is equivalent...

Of course, we should not forget the advantage of the change: now np.arccos(1/result['airmass']) will actually work!

I'm not saying it's not an improvement, but I had a hard time to find which change actually caused this, just by reading the changelogs (first I assumed it's because of the votable unit parsing change in io.votable, but in fact it was due to this (did I mention yet that I really love git bisect?)

@mhvk - is this critical to backport to LTS? There are quite a few conflicts due to the fact #9971 is not included in LTS, so it's a tricky backport and I would prefer to only include it if it's really essential.

@astrofrog - Given that not having #9971 meant that logarithmic units didn't work properly in LTS anyway, I think we should just not backport this one either. Should have seen that at the time....

Ok thanks for the quick reply!

Now seeing #11418, should the changelog entry be moved to 4.2.1 in master too!?