Fail fast on permanent eval sampler API errors

[sylvesterkaczmarek]

Summary

Stop the eval samplers from retrying permanent API failures forever.

Both ResponsesSampler and ChatCompletionsSampler currently put every exception other than BadRequestError into an unbounded exponential-backoff loop. That includes non-retryable API status failures such as authentication, permission, not-found, and unprocessable-request errors.

Fixes #262.

Fix

Both samplers now:

• preserve the existing explicit BadRequestError behavior;
• retry API status codes 408, 409, 429, and 5xx with the existing exponential backoff;
• retry APIConnectionError with the existing exponential backoff;
• immediately re-raise other APIStatusError values;
• allow unexpected non-API exceptions to propagate instead of treating them as rate limits.

Chat Completions still retries its intentional empty-success-response condition through a dedicated internal exception, preserving that existing behavior without making arbitrary ValueError instances retryable.

Regression coverage

Adds tests for both sampler implementations verifying that representative permanent status failures are surfaced immediately and do not invoke time.sleep.

The retry delay and retryable-status behavior are otherwise unchanged.