The system network tests asserted that queued events, disk space, logs,
version, tasks, and updates were empty against a default state that was
already empty. Those assertions passed even with the fetcher body removed,
so they reported coverage that did not exist.
Seeding a stale value first makes the empty response test prove the fetcher
replaces rather than merges, and makes the failure tests prove the update
closure never runs on a non-success response by pinning the exact seeded
value.
The assertion on the deserialized payload in the empty response test is
left alone: it reads the response body rather than app state, so it was
never vacuous.
Dark-Alex-17
ce8c51b02f