The current implementation loads a page of results from the remote query, but when fetching additional pages these might result in an exception when the remote query has expired and garbage collected.

I'm wondering if we should re-incarnate the query again, then scroll to the requested page.
There are two downsides:

• if the user could avoid the remote query to go in timeout that's obviously more efficient
• the scroll result might not be accurate, e.g. if the index has changed in the meantime there might be gaps or duplicates.