[DOC] add newer features to the hierarchical forecasting tutorial - grid search, metrics, ensembles, etc

[fkiraly]

sktime features for hierarchical and panel time series forecasting have progressed a bit since we presented our initial tutorial at pydata Berlin 2022.

It would be nice if some of the newer features were added in "the right places" of the hierarchical forecasting tutorial https://github.com/sktime/sktime/blob/main/examples/01c_forecasting_hierarchical_global.ipynb.

Features I can think of:

• use of the pooling parameter in make_reduction, as was added by recent PR of @danbartl
• native pooling and aggregation features of forecasting metrics (classes)
• get_fitted_params in case of "hierarchical-by-broadcast"
• using grid search tuners such as ForecastingGridSearchCV to tune best models by instance or globally
• use of ForecastByLevel for granular, custom control of hierarchical pooling
• ReconcilerForecaster that @ciaran-g added since then, with more reconciliation strategies
• HierarchyEnsembleForecaster by @VyomkeshVyas, if/once it's ready: [ENH] HierarchyEnsembleForecaster for level- or node-wise application of forecasters on panel/hierarchical data #3905

anything else that should be here? FYI @danbartl, @KishManani, @ciaran-g, @VyomkeshVyas, @topher-lo

I think this is a potential good first issue since each topic could be a small vignette, added in the right place, or modifying one of the existing vignettes. Contributors could pick one topic.

[fkiraly]

relevant recent discussion with @anthonygiorgio97 with some code that might be reducible to nice vignettes:
#4101

[ahmd-nish]

Hi @fkiraly ,
I would like to contribute to this issue. Can I work on it ?

[fkiraly]

Sure! Which topic would you like to work on, @nish-ahmd-it?

[ahmd-nish]

@fkiraly shall i focus on use of the pooling parameter in make_reduction, as was added by recently ?

[Aarthy153]

Hi, @fkiraly, I am willing to work on this issue. Shall I work on ReconcilerForecaster? By the way, can you explain the specific requirement in detail? Thank you!

[pranavvp16]

Hello , @fkiraly , I would like to add the feature of ForecastingGridSearchCV after models like NaiveForecaster and other models are fit into data in the notebook. I think to add this feature after we have received a prediction with default parameters , and then we can use ForecastingGridSearchCV to fine tune the model with optimal parameters

[fkiraly]

@Aarthy153, yes, that would be great!
The requirement is one orw two vignettes, in the right location (possibly next to Reconciler), showcasing its use.

@pranavvp16, that would be great! The focus should be using it in the hierarchical case. Same idea - a vignette somewhere suitable in the notebook.