Apply rows for implicit partial sync with prio 0

[simolus3]

Changes from buckets with priority zero are supposed to be applied even if there's local data (in ps_crud, or a target_checkpoint_request_id that has not yet been reached). Usually, this works with the PowerSync service sending priority 0 data followed by a partial_checkpoint_complete.

When all buckets in a checkpoint have priority zero, there's no partial_checkpoint_complete message. The service sends all bucket data followed by checkpoint_complete. We should still be able to run an implicit partial checkpoint at this point.

Note

We can't simply ignore the can_apply_sync_changes check for full checkpoints consisting exclusively of priority zero buckets. The reason is that a full sync also considers ps_updated_rows, so a concurrent checkpoint before these have been uploaded would lead to flicker even for insert-only tables. Running a partial checkpoint before a full checkpoint (which would then fail due to local data) is necessary.

To fix this, we check if all buckets are of priority zero when a full checkpoint failed due to local data. If that's the case, we retry as a partial checkpoint. The sync client needs to react to this "downgrade", because we shouldn't set the global lastSyncedAt field here.

AI use: Reviewed with Claude Code, which found some issues where a full checkpoint downgraded to a partial checkpoint would not be retried after a crud upload notification. Fixed that with Claude as well.

[rkistner]

There's another case here: If we receive a checkpoint_complete with both priority 0 and priority 1 buckets, and the apply is blocked by local changes - can we still apply the priority 0 changes?

[simolus3]

There's another case here: If we receive a checkpoint_complete with both priority 0 and priority 1 buckets, and the apply is blocked by local changes - can we still apply the priority 0 changes?

Wouldn't we receive a partial_checkpoint_complete for the priority 0 subset first then? That should apply regardless, and I think we should get that message even if there's no new data on the priority 0 buckets.

[rkistner]

Ok, confirmed that the service always sends a partial_checkpoint_complete between buckets of different priorities. We should probably formalize the spec for that, to ensure we don't change that in the future without considering this use case