if weights were None, then it used the model.weights, which was in some cases the wrong shape or the wrong content.
This is broken for WLS if exog is not None, but weights is None.
It is also broken for OLS, because weights in OLS has full length. (refactoring ? IIRC it was scalar 1. before)
This PR fixes the weight handling, but doesn't address the formula support in #987
BUG: fix weight handling, fails in some cases, see #987
needs unit test, but then should be merged soon.
handling formulas in #987 can go into new PR
I don't find any unittests for this. I thought it's tested through influence_diagnostic but messing up the numbers doesn't raise any test failures.
TST: predstd move __main__ to a test module
BUG: wls_prediction_std: fix if weight is scalar
TST: wls_prediction_std: additional tests
TST wls_prediction_std add tests for OLS and confint