Fix locking while writing snapshots to clients

[benbjohnson]

While writing snapshots to connected replicas, we are seeing errors from Rusqlite that suggest read locking is not working properly. This primarily noticeable for larger databases as the client will recover automatically after the snapshot has been written.

RusqliteError { query: "", err: SqliteFailure(Error { code: FileLockingProtocolFailed, extended_code: 15 }, Some("locking protocol")) }