Add smart retry with timeout on startup

[iliia-veselov]

Summary

When SDMX Proxy is deployed as part of a Helm chart, some dependent components
(e.g. Redis, upstream registries) may be temporarily unavailable during startup.
Currently the proxy fails immediately on connection errors. It should instead
retry with a configurable timeout, similar to how statgpt-backend handles
startup dependencies.

Motivation

• The proxy is now deployed via Helm alongside other services that start
  concurrently. Transient unavailability at boot time is expected.
• Immediate failure forces manual restarts or complex init-container
  orchestration that a simple retry loop would eliminate.

Proposed behavior

1. On startup, if a required dependency (Redis, registry health check, config
   endpoint, etc.) is unreachable, retry the connection instead of failing.
2. Retries should use exponential back-off with a maximum total timeout.
3. Both the retry count / interval and the overall timeout should be
   configurable (e.g. via environment variables or application.yaml).
4. After the timeout expires without a successful connection, fail with a clear
   error message indicating which dependency could not be reached.

Reference

• statgpt-backend already implements startup retries and timeouts for both of
  its components -- use it as a reference implementation.

Priority

Nice to have (not critical, but important for production stability).