0369 — The payload generic is named Message, cascading Work* type names to Message*

Context. The producer/consumer generic type param was WorkType, while the datastore files already used Message as their own internal generic param name — an inconsistency, and a vocabulary choice worth researching before v1.

Decision. Message everywhere the param means “the caller’s payload type”. Landscape research: job-queue libraries (River, Oban, Faktory, Sidekiq, Asynq, BullMQ) converge on Job/Task; log/pub-sub systems (Kafka, RabbitMQ, NATS, SQS, GCP Pub/Sub, pgmq) converge on Message. pgmq — the closest same-category Postgres-native project, with zero Kafka lineage — independently landed on Message, which tipped this away from reading as “just copying Kafka”: the architecture (partitioned topics, log compaction, routing-key bindings, fan-out, cursor-based consumption) is log/pub-sub shaped, not job-execution shaped. Cascaded renames: WorkProducer to MessageProducer, NewWorkProducer to NewMessageProducer, WorkConsumer to MessageConsumer, NewWorkConsumer to NewMessageConsumer, WorkConsumerConfig to MessageConsumerConfig — WorkProducer[Message] would have read inconsistently.

Consequences. pkg/concurrency’s own WorkType param (Queue[WorkType]/PressureQueue[WorkType]) was deliberately left alone — its fate was tied to the queue/pool-limiter decision, not this one.