Simplify control flow in sync iteration

[Chriztiaan]

This ports similar changes from the JS SDK (itself a port of powersync.dart#410) to the sync client in the .NET SDK.

Addresses #81.

When a sync stream closed, a few already-queued messages could still be sent to the sync core after it had stopped listening. Those turned into No iteration is active errors that surfaced in crash reporting without the app being
able to catch them.

The iteration now processes everything in one sequential loop rather than alongside a background task, so once a stream ends nothing further is sent. Also uses the core's newer connection command to report connect/disconnect state instead of setting it manually, and makes stream read failures flow into the normal retry path.

AI disclosure

This PR was created with the help of Claude Opus 4.8. Help constitutes assistance in research, planning, and rough outline of implementation. Beyond having a hand in the implementation, I have also manually tested this work.

[LucDeCaf]

The main implementation looks good to me, it mostly matches the quick-and-dirty version I made when I first checked out #81.

I ran this by Opus 5 to see if it could spot anything; it found a few potential races/guaranteed retries which we worked on fixing (it also found a number of issues which didn't actually exist, of course.)

The brief summary is that the socket dying or the iteration ending at the wrong time could cause an exception to be thrown or resources to not always be cleaned up, plus a small regression where CloseSyncStream.hide_disconnect's value is ignored.