We memoize the result object so that when it is later evaluated the same
Elasticsearch result can be used for the `total_count`.

We also extract a shared test examples for the 3 different kinds of
Elastic search results to reduce the chance of regression here.

This memoize approach to solving this problem was originally introduced
in https://gitlab.com/gitlab-org/gitlab/-/merge_requests/13120

Note that some of the methods being memoized have arguments and as such
the memoization is not logically consistent from the code perspective.
We must memoize the results without the page argument because
the current page we're on - if we were to memoize with dynamic
parameters we would end up hitting ES twice for any page that's not page
1, and that's something we want to avoid.  It is safe to memoize the
page we get here because the method is always called before
the equivalent `#_count` method.

Read more at https://gitlab.com/gitlab-org/gitlab/-/issues/219209