[ENH] Parallelize `GAM.gridsearch`

[ankurankan]

The current GAM.gridsearch method sequentially goes over the grid and computes the score. As each iteration of grid search is independent, there is potential to parallelize this loop to make it more efficient.

[fkiraly]

Integration with skpro as in sktime/skpro#634 would solve this, because then you can use the skpro grid search (or any tuning relying on hyperactive) that is parallelized.

Another option might be to use hyperactive grid search to avoid duplication of the parallelization code, i.e., GridSearchSk on a simple evaluation wrapper for pygam.