Use structured concurrency for sync jobs

[simolus3]

The jobs to download and upload data (as well as the one forwarding sync status changes) are all operating independently from each other, which can potentially cause issues stopping these jobs as once.

This adopts a SupervisorJob for the three, allowing us to cancel just one job in disconnect that's guaranteed to get them all.

I've also slightly refactored our test client so that it properly forwards cancellations (when we cancel the sync jobs, that propagates to SyncStream.streamingSyncRequest in the request.execute block). With the mock HTTP engine from ktor we unfortunately have no access to that scope, but by implementing a fairly trivial custom engine we can implement the same functionality while also reporting downstream listeners being cancelled.
This then allows testing that calling disconnect() or close() on a database does indeed close the underlying HTTP response stream.

[stevensJourney]

LGTM