Add test for missing CGKA ops

[alexjg]

This PR is a test which exposes a bug which I'm running into with the CGKA ops. The problem occurs when we sync membership operations and prekey operations but not cgka operations to another peer who initially had no data. Calling Keyhive::cgka_ops_for_agent (or anything which ends up calling Document::cgka_ops) will panic in CgkaOperationGraph::topsort_for_heads because there is an assertion (on line 272 of keyhive_core/src/cgka/operation.rs) that the heads of the operation graph are not empty - but they are empty because we haven't interacted with the cgka graph at all yet.

[expede]

Here's the relevant diff

-        let mut events = alice
-            .membership_ops_for_agent(&bob.active().clone().into())
-            .into_values()
-            .map(|v| StaticEvent::from(Event::from(v)))
-            .chain(
-                alice
-                    .reachable_prekey_ops_for_agent(&bob.active().clone().into())
-                    .into_iter()
-                    .flat_map(|(_, v)| {
-                        v.into_iter().map(|v| {
-                            StaticEvent::from(Event::<MemorySigner, _, NoListener>::from(
-                                Rc::unwrap_or_clone(v),
-                            ))
-                        })
-                    }),
-            )
-            .collect::<Vec<_>>();
+        let mut events = alice.static_events_for_agent(&bob_indi.into()).unwrap();

[alexjg]

The issue here is that I actually want to do the membership and prekeys ops in a separate step to the cgka ops in order to facilitate partial synchronizatipn and prioritization. I.e. I want to sync membership and prekeys ops at the beginning of sync, but then do things like prioritise the list of out of sync documents and sync cgka ops for higher priority documents first.

[alexjg]

I think it's also always possible - even with the static_events_for_agent call - that we fall over half way through syncing ops and end up in the situation where we've processed the membership ops and prekeys ops but not the cgka ops so the cgka opgraph heads could legitimately be empty.

[expede]

Ah okay. I think the problem was that you're not syncing CGKA ops here. Prekeys != Cgka

[expede]

I have a couple quetsions about the sync strategy, but we'll talk live shortly :)

[expede]

I added an assert! below just to make the fix easier to confirm. If this is the behaviour that you expect, delete these comments!

[expede]

events_for_agent here is the same as static_events_for_agent, but without serialization. We use a helper below that can take unserialized events (though it actually serializes under the hood for code sharing reasons but it's more convenient this way IMO but lemme know if you disagree!)

I also switched to using an ingestion helper. ingest_event_table and its sibling ingest_unsorted_static_events are hidden behind a cfg(any(test, feature="test_utils") because they're not super efficient, but they're really helpful for test setup (like here). They run to a fixed point; if the fixed point is empty then you're Ok, but Err if you don't ingest any new ops between two cycles.