Known bug:

1. calibrate and postStratify do not recompute the
   finite population correction. This matters only
   when the sampling fraction is very close to one

2. svydesign should check that `weights` is the right length

3. options(survey.lonely.psu="adjust") does not do what it says in the 
case of svytotal(). Not yet clear whether it does the right thing, though.


4. Should drop unused factor levels from strata, but don't.