Recover from some errors within sync iteration

[simolus3]

This adds the can_retry method to the new PowerSyncError type (#96). Then, in the sync iteration, we catch errors for each sync line and check whether the error is fatal or whether we allow retrying the operation.

When retrying isn't necessary, we keep the sync iteration future active. powersync_control will still return an error, but the invocation can be retried after e.g. unlocking a database or waiting for a WA-sqlite VFS to settle.

To implement this in a sound way, handling sync lines is split across two functions:

1. StreamingSyncIteration::prepare_handling_sync_line: This method is responsible for writing data to the database, but it does not mutate the internal sync client state. It returns a SyncStateMachineTransition, which describes pending changes to be applied if the call has succeeded.
2. StreamingSyncIteration::apply_transition then updates the internal client state.

The reason for this split is that it makes it easy to argue that we cannot possibly update in-memory state before completing writes to SQLite (thanks Rust!). This ensures that handling sync lines is idempotent wrt. underlying SQLite failures, the property we need.
This also adds tests to block the database in different stages of the sync iteration.

The base branch was changed.