AsyncDBClient uses the reactor to schedule the cancellation handling callbacks, including when the CancellationSource goes out of scope without having been cancelled. This results in a task per network operation being scheduled on the reactor, most of which do nothing. It appears this results in a 95p latency regression in high-concurrency workloads. We should update these to only schedule callbacks on the reactor when an actual cancellation occurs.