Preserve physical row order during deduplication from incremental sync

[emilklindt]

This pull request fixes an issue where the target fails to correctly deduplicate rows that were extracted in succession, due to identical _sdc_received_at and _sdc_extracted_at timestamps. The solution is to preserve physical / stored row order by sorting directly on columns, instead of using the COALESCE function.

Example

When using COALESCE in the ORDER BY, rows with the same values are considered identical, and does not preserve natural sort order descending.

Before Fix (COALESCE used for deduplication)

id	_sdc_received_at	_sdc_extracted_at	Value	Selected row
1	2024-10-02T12:00:00.000Z	2024-10-02T12:00:00.000Z	Oldest	Selected
1	2024-10-02T12:00:00.000Z	2024-10-02T12:00:00.000Z	Older	Ignored
1	2024-10-02T12:00:00.000Z	2024-10-02T12:00:00.000Z	Current	Ignored

After Fix (direct sorting on columns)

id	_sdc_received_at	_sdc_extracted_at	Value	Selected Row
1	2024-10-02T12:00:00.000Z	2024-10-02T12:00:00.000Z	Oldest	Ignored
1	2024-10-02T12:00:00.000Z	2024-10-02T12:00:00.000Z	Older	Ignored
1	2024-10-02T12:00:00.000Z	2024-10-02T12:00:00.000Z	Current	Selected

Impact Assessment

This is not a breaking change for most sources, as long as they consistently include or exclude the _sdc_extracted_at value. If only some rows have this timestamp, those rows will be prioritized, but this scenario is unlikely.