retry on transient failures

[korbinib]

• retry on transient failures (HTTP 429/500/502/503/504 and connection-level errors like timeouts/dropped connections), but do not retry definitive errors like 404.
• Polite backoff
• User agent
• AUTOCOMPLETE_MAX_RETRIES env var (default 4)

trade-off: if one of the API endpoints is entirely down there is now ~15s retrying before giving up, vs. failing fast before - considered acceptable, since we run this asynchronously

---

📚 Documentation preview 📚: https://databank--513.org.readthedocs.build/

[mdondrup]

This looks like a robust improvement. I think we may need to reconsider timeouts and retries in case it doesn't work. On the other hand, I think it isn't necessary to overthink it and adapt and fine-tune our service to compensate for the deficiencies of another.