ENH: hurdle count models - joint estimation and common parameters

[josef-pkt]

current version of hurdle model in #7973 use separate estimation.

R pscl has option for joint estimation

One use case for joint estimation will be when we want to impose common parameters across submodels (when we use the same family in both).
example in negbin-negbin, we could impose common dispersion

related usecase would be penalized estimation where we penalize towards identical parameters in both models.

We also still need the pieces for joint estimation like score_obs, hessian for robust cov_types. Those are also not available in the PR version.

[josef-pkt]

We need this also when not all parameters are identified in zero model
e.g. #7973 (comment)