identical to #1734 except rebased and resolved merge conflicts.
see 1734 for description and links to related issues, partially closed
Added missing offset to starting value calculation
Rename eta to lin_pred
Close second part of issue #1720
Rename Y as endog
Handle maxiter=0 as a special case
Added _offset_exposure to _data_attr (but still fails in test_shrink_…
Increase pickle size limit (was failing on GLM)
Break early in IRLS if convergence is detected
Coverage increased (+0.01%) when pulling 4a8a529 on josef-pkt:glm-maint_1734_rebased into f7483fd on statsmodels:master.