GLMResults inconsistent distribution in pvalues

[josef-pkt]

pvalues still use the t-distribution, conf_int use the normal distribution
Looks like some incomplete refactoring that escaped us.

There is currently no way to overwrite which distribution is used.

also summary show t not z (needs use_t=False)

[josef-pkt]

I think the switch from t to normal for GLM was this
4594537

however, GLMResults has it's own pvalues method based on t, that overwrites the "generic" normal case.