The redis transport lane inherits the mock config's 10s assertion budget and
2 CI retries, both of which were tuned against the in-memory stream store.
Every stream event here crosses a real Redis round-trip, and the pause/resume
and rehydrate scenarios replay an entire job, so the same waits run much
closer to their budget than the memory shards ever do.
The suite is serial (`workers: 1`), so this is per-operation latency rather
than worker contention — which is why this lane reports flaky runs the memory
shards do not.
Raise the default assertion budget to 20s and allow one more retry for this
config only. The memory shards keep their current settings.
