Improve service correctness, operability, and testability for production deployments

[darthfork]

Summary

A few infrastructure and correctness details will matter more as the service is deployed more widely: API status metrics are currently fragile, graceful shutdown is missing, and the metric/registry setup is tightly coupled to global state.

This issue tracks foundational cleanup that improves operability, testability, and production readiness.

Why this matters

• Incorrect HTTP status recording undermines observability.
• Lack of graceful shutdown can drop in-flight requests during deployments.
• Tightly coupled global metric registration makes testing and composition harder.
• Logging on the hot path can create unnecessary I/O overhead.

Goals

• Improve correctness of service-level observability.
• Make the service friendlier to orchestration environments.
• Improve internal structure for testing and future extension.

Suggested scope

• Fix status recording middleware.
• Add graceful shutdown behavior.
• Reduce per-request log verbosity on normal paths.
• Refactor metric registration away from hard-coded global defaults where practical.

Child issues

• Fix HTTP status recording in middleware.
• Add graceful shutdown and worker draining support.
• Rework metric registration for better testability and composition.
• Reduce hot-path request logging overhead.

Acceptance criteria

• Service-level request metrics accurately reflect response codes.
• The process shuts down cleanly during deploys and restarts.
• Tests can use isolated registries without global coupling.