Your solution #1 is very frustrating to me as well, for a number of reasons.  Simply put, we have to be able to compare incoming data to target data for normal ETL operations. 

One way around this is to create a view of your target silver table, outside of your pipeline (this part is key), and compare against that view.  The pipeline will give you a warning telling you that you're being naughty, but it will let you do the comparison anyways.  We have had to do this and it does work, although it's clearly not a desirable solution to have to maintain this set of views, but it would allow you to perform the deletion check.

Example: `create or replace view silver.vw_target_table as select * from silver.target_table;`

Someone else might be able to chime in with a more elegant solution, but that's what I've got for you at least!