BUG/REF: redundant code in genmod get_prediction_glm, ignores link

[josef-pkt]

update It looks like there are some bugs, some options seem to be ignored,
e.g. I don't see any offset handling in the code for endpoint transformation confint, it just uses linpred = x b.
looks like false alarm
more in later comments

The get_prediction_glm function in genmod has weights handling copy pasted from regression linear model version.
weights are not used in GLM.
GLM and GLMResults predict do not support weights, no freq_weights or var_weights. Maybe those should be added.

The function needs exog everything else we could delegate to results predict instead of using model predict.
The exog need to go through transform handling.
In GAM we also need creating the basis function for the predict exog.

[josef-pkt]

link argument is ignored, instead the attached model.family.link is used.

[josef-pkt]

offset handling, reading code

predicted mean looks correct it goes through model.predict with keywords
predicted_mean = self.model.predict(self.params, exog, **pred_kwds)
PredictionResults uses the linear prediction result only in EP confint

regression linear PredictionResult also goes through model.predict, so pred_kwds like offset are taken into account
predicted_mean = self.model.predict(self.params, exog, **pred_kwds)
confint for linpred is predicted mean +/- crit * std. So this is also correct.

So offset handling in get_prediction looks correct
I don't see unit tests for it.