Documents the scenario raised in review: when the receipt stream has
several writers and the worker doing the paging has heard from one
writer further than another, get_linearized_receipts_for_all_rooms
bounds its SQL on the highest position in the token and filters rows
from the lagging writer afterwards, so the returned token can claim
that writer's receipts as handled even though they were never returned.
This test is expected to fail until paging filters per writer in SQL
and returns a per-writer token.