fix: Do not await for responses when recovering pub filters

[msfstef]

The recover_shape and refresh_publication are in the init path of ShapeCache, and there is no reason why we would be waiting for responses.

The recover_shape has no expected failure as it simply updates a map with appropriate counters.

The refresh_publication will schedule a publication update, but we don't have to wait for it to finish to consider ShapeCache ready, as the shape consumers are already up and running and any new shapes will trigger another publication update after this one.

These changes are based off of Sentry errors noticed from the cloud.

I've also set the publication update debounce time to 0ms, so now it simply waits until end of current process message queue. This addresses an unnecessary delay in prepare_table calls (@robacourt), since even when there is no contention it would still debounce for 50ms. I think that simply reinserting the update at the end of the message queue should still handle things pretty well but we can see the benchmarks.

[msfstef]

benchmark this

[github-actions]

Benchmark results, triggered for a8a1a

• concurrent shape creation completed

• diverse shape fanout completed

• unrelated shapes one client latency completed

• write fanout completed

• many shapes one client latency completed

[robacourt]

Nice work! I've just left one comment