Support safe config reload via SIGHUP and /-/reload #1390

Proposal

Support reloading the postgres_exporter configuration without restarting the process, following Prometheus' lifecycle behavior:

• SIGHUP triggers a configuration reload.
• POST /-/reload triggers the same reload path.

Use case

Operators need to rotate auth_modules and other reloadable file-backed configuration without restarting the exporter. This is particularly useful with Kubernetes Secrets, external secret managers, and multi-target /probe deployments.

Several old pull requests attempted parts of this:

• Add support for dynamic reload on config change #361 added SIGHUP and automatic watching of DATA_SOURCE_URI_FILE, but predates the current runtime architecture and multi-target support.
• Add config reload via /-/reload #821 and Add config reload via /-/reload #822 added /-/reload, but have become stale.

Expected behavior

1. Parse and validate the candidate configuration before applying it.
2. Apply a successful reload atomically.
3. Preserve the last-known-good configuration after a failed reload.
4. Return a non-2xx response from /-/reload when reload fails, including a useful error message.
5. Handle concurrent or repeated reload requests safely.
6. Update the existing config reload success/timestamp metrics consistently for both HTTP and SIGHUP reloads.
7. Document the signal, endpoint, and failure semantics.
8. Add tests for successful reload, invalid configuration, signal handling, and concurrent requests.