BUG: Vasp setups

It appears that in implementing the new xc and pp syntax I introduced a small bug which breaks the specification of alternative setups. Here is a fix.

The root cause of the problem is that I did not understand how the setups were supposed to work due to the lack of documentation. I have therefore tried to also address this by adding a section to the documentation of the Vasp calculator. It would probably be wise if somebody familiar this feature could have a look and check that it is correct!

I note that @jkitchin is moving to a different syntax in his fork at https://github.com/jkitchin/vasp which uses lists instead of dictionaries, so perhaps there is a wider discussion to be had about this syntax. For now though I propose that we undo the regression...