Update annotations after label removal

[leoll2]

When the user removes a label, the existing annotations should be updated accordingly by removing that label where present. If a shape remains without a label, that shape should be removed as well. Likewise, if an image remains without shapes, it becomes empty-labeled (annotation_data=[]). For tasks that don't support empty label, namely multiclass classification, set it as unannotated (annotation_data=None with user_reviewed=false).

The dataset_items_labels table can be used to efficiently find the annotated media that should be updated.

For simplicity, the annotation update can be done synchronously with the project labels update request, even if this may increase the endpoint latency when the dataset is large. We can optimize this at a later time.