Existing solutions / potential approaches

• nf-cascade (daisy-chaining)

◦ No resume, no error handling, poor parallelisation

Hey, that's not true. You can resume, and parallelisation is handled by the executor as normal.