Support programatic leap seconds

Currently, there’s no way for astropy to programmatically change what the leap-seconds are, because the leap seconds are accessed through ERFA. This is problematic because it means when there’s a new leap second we have to do a bugfix release.

The fix in Astropy itself would be pretty easy, as it can be extracted from the IERS files that we already are using for DUT1 and such. But the non-trivial part is updating ERFA to have an interface for setting the leap second - see liberfa/erfa#43 for discussion on that.

cc @mhvk

[eteq]

At the astropy coordination meeting 2018, @adrn, @timj, @stargaser, et al. decided this is a good thing to prioritize for 3.2 (which means working on the liberfa fix asap). Hence the milestoning above.

[mhvk]

@bsipocz - no, not yet, we now have in principle the ability to update the leap second table, but we still need to put in place the machinery to get that updated table.