Improve retry/backoff logic on "backend internal errors"

[kusor]

In case of job's tasks, we could keep retrying to save when a BackendInternalError is raised. The job will continue if we can write tasks results to the backend before it times out and, otherwise, it will timeout as expected.