observability-sdk-validation-langfuse-sdk-throws-pydantic-v1-validation-errors--39027f68

THEN Upgrade to langfuse >=2.57.9 if not already, and ensure no missing fields in usage details. As a workaround, preprocess the OpenAI response usage object by adding default 0 values for fields like `audio_tokens`, `cached_tokens`, `reasoning_tokens`, `accepted_prediction_tokens`, `rejected_prediction_tokens` before passing to the Langfuse update method.

def safe_usage(usage):
    for keys in [('prompt_tokens_details',), ('completion_tokens_details',)]:
        obj = usage.get(keys[0], {})
        if not obj:
            usage[keys[0]] = {'cached_tokens': 0, 'audio_tokens': 0}
        else:
            obj.setdefault('audio_tokens', 0)
            if keys[0] == 'completion_tokens_details':
                obj.setdefault('reasoning_tokens', 0)
                obj.setdefault('accepted_prediction_tokens', 0)
                obj.setdefault('rejected_prediction_tokens', 0)
    return usage