Partial update of records will set all non-PK columns to null

[Tolsto]

Describe the bug
When using the Postgres tap with log-based replication a delete will generate a record message that only contains the primary key properties and metadata columns like _sdc_deleted_at. This will cause the Snowflake target to reset all other columns to null as the MERGE query does no distinguish between missing columns in the record message and columns that are truly null.

This behaviour doesn't matter much when you are using hard-delete and are going to delete the row in the target anyway. However, when you actually want to keep deleted source data in your DWH then it's a big problem.

To Reproduce
Steps to reproduce the behavior:

1. Use the target with a tap that sends incomplete record message (e.g. Postgres with log-based replication when doing deletes)
2. Generate an incomplete record message for an existing row in the target.

Expected behavior
A partial update should not set unaffected columns to null

Your environment

• Version of target: 2.1.0
• Version of python: 3.9.12