fix(connectInfiniteHits): fix page state when adding or removing widgets

[yannickcr]

Summary

IFW-980

This PR fix multiple pagination reset issues with InfiniteHits:

• When requesting the SearchParameters from the InfiniteHits widget the page was always reset to 0 if showPrevious was false. We now return the original searchParameters (like it was the case before feat(infiniteHits): support new routing system #4040).
• When comparing the currentState with the prevState (to know if we need to reset the widget internal cache) we now exlude hierarchicalFacets, disjunctiveFacets and filter hierarchicalFacetsRefinements and disjunctiveFacetsRefinements content to only keep the actual refinements. This is very hacky 🤢 . Any idea to make this cleaner and/or more reliable is welcome 😉 .

[Haroenv]

I don't like this, because as in React InstantSearch, it doesn't work properly with refresh. I have no clue what a solution can be.

An alternative would be:

1. always request all the pages shown now
2. split up caching for each request
3. read from cache, instead of saving in widget itself

This requires the client v4 and maybe even more though, but would make the design better IMO, since there would be no need to do these conditions, since they're done by the cache level on the client.

[samouss]

It doesn't have the expected behavior anymore. The widget has to set a default value otherwise it gets controlled by its parent. Here is an example of a structure that wouldn't behave like expected:

search.addWidgets([
  hits(),
  pagination(),

  index().addWidgets([
    infiniteHits(),
  ]),
]);

Here is the live example (CodeSandbox doesn't work with the ES version at the moment). You can select a page inside the pagination it shouldn't impact the infiniteHits but it does. The widget doesn't have a default value anymore. We have to keep the page of 0 no matter what when the widget is mounted without refinement.

[samouss]

I think we forgot to handle facets & facetsRefinements.