Terraform: managed cluster, remote state and reproducible rebuild

[FullFran]

Everything up to here is applied by hand. This issue makes the infrastructure
reproducible.

Scope

• Terraform for a managed cluster and its node pool
• Managed Postgres and Redis, or an explicit written decision to run them
  in-cluster with the trade-off stated
• Remote state with locking — never local state
• Secrets sourced from the cloud provider's secret manager, not from .tfvars
  in the repository
• Separate dev and prod workspaces or directories
• terraform destroy followed by a full rebuild from zero, to prove it is
  genuinely reproducible
• Document the measured monthly cost, and how to bring it to near zero when idle

Notes

The rebuild-from-zero step is the whole point. Infrastructure code that has never
been destroyed and recreated is a guess, not a guarantee.

Record the real cost. The reason to know it is that the crossover analysis issue
depends on it.