BUG GenericLikelihoodModel does not propagate kwargs in __init__

[josef-pkt]

I ran into the bug with hasconst that was not propagated
see #7035 (comment)

There might be other keywords that should be propagated, e.g.
I guess that the data handling treatment for extra arrays like offset does not work, which, I guess, breaks conversion to ndarray and inclusion in missing value handling.

currently GenericLikelihoodModel.__init__ just attaches everything self.__dict__.update(kwds)

However, if I just propagate all kwds to the super call, then unit tests for miscmodel TLinearModel fail.

This needs review and better separation of which keyword should be propagated to super.

partial fix in #7035: I explicitly propagate hasconst, but leave other kwds as it is currently done.