Fix a race condition around OPFS databases

[simolus3]

In web_sqlite_open_factory.dart, we create a shared mutex derived from the database path when a database is not accessed through shared workers. This is relevant when serving Flutter apps with COOP + COEP headers, because we're able to use a VFS implementation based on OPFS and dedicated workers in that case (so no shared worker is involved).

Unfortunately, web_sqlite_database.dart allocates its own (anonymous) mutex and passes that as an option to the open factory. This means that the shared mutex isn't used, and two tabs could race to use the database simultaneously, which may cause errors as the VFS runs into blocked sync access handles.

Note: A side-effect of this change is that databases with shared workers have no per-tab/client-side lock. This shouldn't be an issue because the shared worker is supposed to hand out leases as well, but it's something to be aware of. This is another improvement though:

• Previous behavior: Two tabs have an independent mutex on a connection backed by a shared worker. They can't actually run statements concurrently, but they'd be able to both obtain an exclusive context in parallel 😬
• New behavior: There's only a single mutex, managed by the shared worker. So two tabs can't obtain an exclusive context at the same time, but there's no timeout for locks.