Kubeflow run fails when downloading models with "too many requests"

With a high number of documents and low batch size resulting into a potential high number of parallel pods, models are downloaded concurrently and it creates a _too many requests_ error, which results in failures.

As pointed out by @SantanaTiago
> Regardless if batch size fix this issue or not, I think some kind of retry mechanism to get models (init Docling converter) should be implemented