so there's no way for retries to be configured to always restart from failure?

You can use run_retires -

FROM_FAILURE

https://docs.dagster.io/deployment/run-retries#retry-strategy

Wanting to be able to set that globally (instead of per job) seems like a reasonable request to me though