adjusted tests and added tests to @bashtage PR #1251 , supersedes now that PR
open problem: rsquared doesn't match, see #1252 (not for this PR, since I don't know how #1252 is supposed to be "fixed")
llf, aic and bic differ now from Stata, see and closes #1170
ENH: WLS log-like now matches GLS for comparable weights
TST: adjust test WLS llf, aic, bic because of different definition
TST add test comparing WLS, GLS, rsquared differs see #1252
Coverage remained the same when pulling 8e08fa6 on josef-pkt:WLS-llf-fix into cc04669 on statsmodels:master.
I think this is ready to merge, TravisCI is green