T&D Queries should analyze less data where possible by only considering recently emitted rows

[evantahler]

Datawarehouses bill based on how much data is analyzed in the query. If we can add more WHERE clauses to limit which rows are considered, that will help.

See https://github.com/airbytehq/typing-and-deduping-sql/pull/23/files for some ideas of what to do. Also, read this slack thread.

Add a test case for this:

• a sync crashing and having un-typed data from a previous run

[edgao]

potentially relevant past issue about legacy normalization partition pruning #14070

[evantahler]

^ tldr - save the results of MAX(foo) queries so we don't have to ask it over and over again

[edgao]

putting this down for now to focus on async standard inserts. there's a loose pr where I sketched out the interface diff (#31191), but I didn't get time to modify/add any test cases for this.

[evantahler]

@edgao I think you started this story perhaps?