Partition level lineage tracking

When appending data to an existing data source Spark creates a new partition. Even though appended partitions should have the same schema the way how the data was computed (the lineage) could be different for every partition. Tracking lineage information at a partition level is useful for cases when partitions are subsequently read separately. So the derived lineage could be tracked more precisely by avoiding tracking the lineage of the partitions that were not actually used for subsequent computation.