When LLM translator returns invalid response and we just retry the same request on low temperature, we got the same response and as result we have always the same (incorrect) response and just spend money in the black hole.

Let's implement more efficient algorithm to retry invalid results.