Fix DNS subject (fixes #265)

[paulgb]

Previously, all DNS records for a given cluster went to the same subject. The cli fetches DNS records using DeliverPolicy::LastPerSubject, so it only gets the last message sent to a subject (so one per cluster). This fixes the issue by adding the backend ID to the subject.

Leaving this as a draft for now because this changes the NATS API between the drone and controller and I haven't thought through a zero-downtime migration path as we try to do for NATS API changes.

Currently, migration requires:

• Re-installing the CLI (cargo install --path=cli in root of the repo)
• Deleting the old stream (nats stream delete dns_record)
• Re-building and running the controller and drone (docker compose build followed by docker compose up)

[paulgb]

Stepped deploy approach:

• Deploy controller. The dns_record_tmp stream will be created, but messages published will appear on both the new dns_record_tmp and dns_record streams.
• Drain and redeploy drones. Drones will produce new messages which appear only in the dns_record_tmp stream, but this is fine because consumers of dns_record have been removed.
• Delete the dns_record stream in NATS.
• In a follow-up PR, change dns_record_tmp back to dns_record.
• Deploy.
• Delete the dns_record_tmp stream in NATS.

[paulgb]

I've decided to roll this into the ongoing state machine refactoring work, so I'm going to close this PR.

[paulgb]

I've decided to roll this into the ongoing state machine refactoring work, so I'm going to close this PR.