RealtimeTrigger has no backoff/retry-limit on repeated connection failures, leading to worker thread exhaustion and instance crash

Describe the issue

RealtimeTrigger (io.kestra.plugin.debezium.oracle) has no backoff or retry limit when the underlying Debezium connector repeatedly fails to establish a connection. Each failed attempt immediately spawns a new worker thread to retry — roughly once per second, with no delay and no cap on concurrent workers. This leads to unbounded resource consumption and eventual instance crash.

To Reproduce

1. Configure a RealtimeTrigger (io.kestra.plugin.debezium.oracle) with
   connection parameters that reliably fail (e.g. invalid driver
   registration, unreachable host, or invalid credentials)
2. Enable the trigger
3. Observe worker thread creation in the Kestra logs:
   docker compose logs kestra | grep AsyncEmbeddedEngine

Expected behavior
Reconnection attempts should back off (e.g. exponentially) and/or be capped at a maximum retry count, to avoid unbounded resource consumption when the underlying connection issue is persistent.

Environment

Environment

Kestra Core: v1.3.30
Plugin: plugin-debezium-oracle 1.4.8
Deployment: Docker (docker-compose), standalone mode

Start at RealtimeTrigger (io.kestra.plugin.debezium.oracle) and inspect the AsyncEmbeddedEngine retry path. Reproduce the failure with invalid connection parameters and monitor docker compose logs kestra | grep AsyncEmbeddedEngine. Done means repeated connection failures no longer cause unbounded worker-thread growth, with the retry behavior verified under persistent failure.