Problem

For TQE, we already have a setting to retry failed LLM requests:
runtimeOptions->plugins->OpenAI->qualityScoreSegmentCheckMaxRetries
We need the same behavior for batch pre-translation.

Solution

Add a system-level config that defines how many times a failed LLM request should be tried again during batch pre-translation.
In the best case, this retry logic should be handled in the lower-level LLM client. Then both TQE and batch pre-translation can use the same retry logic for failed LLM requests.