the messages are not 100% a request-response pattern -> input and output messages are better
Also, the basis for all input and output messages is "input_message" and "output_message" - so they are nicely composed, what about using a similar scheme in flowd-rs?
TODO what about generating the dataclasses in flowd-rs from the schema.yaml? Would that make sense?
remember serde annotations, Rust type mapping, methods that are needed for flowd-rs, relationships and re-use of certain structs (dont know if a generator is so intelligent), code bloat etc.