Add celestial conversion check for different epochs of observation

Like @dsberry said in https://github.com/astropy/coordinates-benchmark/pull/4#issuecomment-10753477 :

> 2) In the table, for the "Epoch of observation", you probably want to use
> the same value (J2000) for all systems, at least initially. Having said
> that, you have a problem if any of the coordinate packages do not allow the
> epoch of observation to be specified. For instance, I'm not sure what the
> current implementation of astropy coords assumes about the epoch of
> observation. I understand Erik has agreed that an ObsTime property is
> needed, but I presume he hasn't yet added it.
>
> 6) A further development could be to add more tag names for other
> combinations, like "fk5_J2010_J2012.34" meaning equatorial fk5 with respect
> to equinox of J2010, observed at J2012.34. Converting to and from systems
> with non-default values for equinox and epoch should test more bits of the
> code. Although of course the total number of tests could become difficult
> to handle.

I agree we should keep the number of tests small.
We could just test `astropy` (once obstime is available) against `pyast` and I don't think we need all systems here.