Improve tab close detection for shared sync worker

[simolus3]

Currently, sync workers detect closed tabs by listening on database.closed for a database backed by sqlite3_web (which internally uses navigator locks to detect closed tabs).

This kind of works, but can lead to deadlocks when a tab is closed while it's serving a request (like a fetchCredentials or requestDatabase call) that would never complete.

To improve reliability, this adopts the usual navigator locks pattern to our WorkerCommunicationChannel as well:

1. Before requesting a sync iteration, tabs acquire a randomly-named lock, and they include that lock name in the request.
2. Like before, the shared sync worker will mark a tab as closed once this lock is available to it. But starting from this PR, it also throws for all pending requests (they would never complete before).

I've also added a small test for the shared sync worker in sync_worker_test.dart. It emulates a shared worker by manually passing ports from a MessageChannel to it. I've also tested this in a demo app (and actually found an issue in sqlite3_web, I'll fix that too).

[Chriztiaan]

I found the following issue which I am not sure if it's a regression:

Launch supabase-todolist demo for web in two chrome tabs. Both show connected state, changes to the source database sync down to both tabs. I refresh the first tab, it never reaches a connected state and the second tab still shows connected. Changes to the source database don't sync down to either tab. If I close the first tab and refresh the second tab the data does then sync down.