Currently, the SUP retries failed fetch requests to MW API with a linear backoff, that increases by 1s per retry. That might make sense in case the API is not available for some reason, but in case of a 429 response (> rate limit enforced by envoy), it would block unnecessarily long. Since envoy rate-limits requests based on a sliding window, we might retry more rapidly.

Since flink's API for async operator retries does not expose the retry cause for calculating the backoff, should implement this at HTTP client level.

AC:

• Both, sync and async client use a retry strategy tailored to retry only 429 with a constant 200ms backoff