fix: only register completions handler when server supports completions

[iqdoctor]

No description provided.

[punkpeye]

Looks like the checks are not passing.

[iqdoctor]

pnpm test

 Test Files  6 passed (6)
      Tests  67 passed (67)
   Start at  08:27:15
   Duration  8.92s (transform 519ms, setup 0ms, collect 1.42s, tests 17.62s, environment 1ms, prepare 732ms)

The runner is failing on InMemoryEventStore > stores events and replays them after a specific event ID.

Locally we always get 3 replayed events because each storeEvent() call falls into a different millisecond. On GitHub Actions, multiple events share the same Date.now() value, so the generated eventIds differ only in the random suffix. Since replayEventsAfter() sorts by the eventId string (see src/InMemoryEventStore.ts#41-67), one of the “future” events can end up ordered before the lastEventId, which means it never gets replayed once foundLastEvent flips to true. The test then sees only 2 events.

We need to make ordering deterministic when timestamps collide—e.g. track an incrementing counter/order column or append it to the eventId—so all events after lastEventId always sort after it, even if Date.now() returns the same value. Once we guarantee insertion order, that test will be stable in CI.

[iqdoctor]

Should I update the tests, or would you prefer a deterministic ordering fix inside InMemoryEventStore (e.g., monotonic counter appended to the event IDs) so CI never drops an event?

[punkpeye]

would you prefer a deterministic ordering fix inside InMemoryEventStore (e.g., monotonic counter appended to the event IDs) so CI never drops an event?

That seems like a more robust approach. I appreciate the contributions!

[punkpeye]

Could you please update the tests?

[iqdoctor]

Make InMemoryEventStore event IDs monotonic and update tests

• generate IDs as ${streamId}_${timestamp}_${counter}_${random} with base36 counter + random tail
• document the new monotonic scheme in InMemoryEventStore
• adjust deterministic-ordering test to validate timestamp/counter/random segments
• ensure counter reset is covered when the millisecond tick advances