rabbitmq:rpc-server lacks --memory-limit and graceful signal handling

[qwant50]

An RPC server is a long-running consumer process like any other, but rabbitmq:rpc-server is missing the process-lifecycle controls that rabbitmq:consumer and rabbitmq:multiple-consumer have:

• no --memory-limit option — servers with slow memory leaks eventually crash with an allowed-memory-size error instead of exiting gracefully so a supervisor can restart them (previously reported in Add --memory-limit to rabbitmq:rpc-server #712 and Memory leak or something else ? #545, both closed by the stale bot);
• no signal handling — SIGTERM/SIGINT kill the process mid-message instead of finishing the current reply first (problematic under supervisord/systemd/Kubernetes);
• no graceful_max_execution / idle_timeout support in the rpc_servers config section.

In #712 a maintainer said a PR would be welcome, and #524 implemented this approach but was also closed by the stale bot before review.

I have a PR ready that re-implements #524 on current master, fixes several defects it had, and adds tests. Submitting it right after this issue.