For cloud databases, in your use case do you care that the queries actually execute or only that they parse? I with every cloud service had their own version of LocalStack/Moto, and while not perfect they've really helped to improve test coverage.

Unfortunately they have to actually execute and have identical semantics to the target source

I recall seeing a project based on DuckDB parser and engine which emulated dialects for one of these but because the sortinh and other behavior is different it wasn't usable