0368 — The consumer shutdown hook is deleted; the app that constructs the datastore closes it

Context. Consume’s exit ran DefaultShutdownFunc, which called ConsumerDatastore.Shutdown and closed the shared pgx pool — after any Consume returned, every other producer or consumer on that PostgresDatastore got “closed pool”.

Decision. The whole hook was deleted after surveying the field (River, gue, pgmq, Pub/Sub, NATS, franz-go, confluent, sarama, amqp091, Watermill, net/http, gRPC, asynq, Temporal): no library has a resource-owning shutdown callback, and every library taking an injected pool leaves closing it to the app that constructed it. Deleted: ShutdownFunc/DefaultShutdownFunc/WithShutdown/WithShutdownTimeout, ConsumerDatastore.Shutdown, the dead Drain(ctx, wg), and Config.ShutdownTimeout — once the hook died the knob bounded nothing live; wind-down is already capped per message by WorkTimeout + WorkTimeoutGrace hard-abandon. PostgresDatastore.Shutdown() was renamed Close(), matching what sql.DB/pgxpool call the app-owned teardown.

Consequences. Ownership rule for both components identically: the app that constructed the datastore closes it (defer ds.Close()); producers and consumers borrow it and never close it. The Go team’s rejection of ctx-shutdown for net/http (golang/go#52805) validates the model: their objection is that ListenAndServe returns before the drain; Consume blocks through the drain and returns nil — the Pub/Sub Receive/asynq/Temporal camp.