Is there anything to make aws respect change sets so that you can always roll forward? The headache is that if a aws cloudformation change set fails, you get stuck in some state and you have to delete the stack to try again. Which isn't ideal. Then when you go to delete the stack, that too can fail because the stack can't recursively delete things like s3 buckets. Does terraform improve on this?

yes

most of the time you can: • re-apply and TF will resolve the changes and get you into correct state • you can taint selected resources to recreate them after next plan+apply cycle

Why hasn't aws done this? i'm baffled, do they just not care to match expectations with modern devops functionality?

Not sure, never used CloudFormation to be honest, so I'm kinda surprised it doesn't have that kind of a feature